1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
35 #include <linux/interrupt.h>
37 #include <linux/tcp.h>
38 #include <linux/sctp.h>
39 #include <linux/pkt_sched.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/prefetch.h>
48 #include <scsi/fc/fc_fcoe.h>
51 #include "ixgbe_common.h"
52 #include "ixgbe_dcb_82599.h"
53 #include "ixgbe_sriov.h"
55 char ixgbe_driver_name[] = "ixgbe";
56 static const char ixgbe_driver_string[] =
57 "Intel(R) 10 Gigabit PCI Express Network Driver";
59 char ixgbe_default_device_descr[] =
60 "Intel(R) 10 Gigabit Network Connection";
62 static char ixgbe_default_device_descr[] =
63 "Intel(R) 10 Gigabit Network Connection";
68 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
69 __stringify(BUILD) "-k"
70 const char ixgbe_driver_version[] = DRV_VERSION;
71 static const char ixgbe_copyright[] =
72 "Copyright (c) 1999-2012 Intel Corporation.";
74 static const struct ixgbe_info *ixgbe_info_tbl[] = {
75 [board_82598] = &ixgbe_82598_info,
76 [board_82599] = &ixgbe_82599_info,
77 [board_X540] = &ixgbe_X540_info,
80 /* ixgbe_pci_tbl - PCI Device ID Table
82 * Wildcard entries (PCI_ANY_ID) should come last
83 * Last entry must be all 0s
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
88 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
98 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
99 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
100 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
117 /* required last entry */
120 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
122 #ifdef CONFIG_IXGBE_DCA
123 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
125 static struct notifier_block dca_notifier = {
126 .notifier_call = ixgbe_notify_dca,
132 #ifdef CONFIG_PCI_IOV
133 static unsigned int max_vfs;
134 module_param(max_vfs, uint, 0);
135 MODULE_PARM_DESC(max_vfs,
136 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63");
137 #endif /* CONFIG_PCI_IOV */
139 static unsigned int allow_unsupported_sfp;
140 module_param(allow_unsupported_sfp, uint, 0);
141 MODULE_PARM_DESC(allow_unsupported_sfp,
142 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
144 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
145 static int debug = -1;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
149 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
150 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
151 MODULE_LICENSE("GPL");
152 MODULE_VERSION(DRV_VERSION);
154 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
156 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
157 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
158 schedule_work(&adapter->service_task);
161 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
163 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
165 /* flush memory to make sure state is correct before next watchdog */
166 smp_mb__before_clear_bit();
167 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
170 struct ixgbe_reg_info {
175 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
177 /* General Registers */
178 {IXGBE_CTRL, "CTRL"},
179 {IXGBE_STATUS, "STATUS"},
180 {IXGBE_CTRL_EXT, "CTRL_EXT"},
182 /* Interrupt Registers */
183 {IXGBE_EICR, "EICR"},
186 {IXGBE_SRRCTL(0), "SRRCTL"},
187 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
188 {IXGBE_RDLEN(0), "RDLEN"},
189 {IXGBE_RDH(0), "RDH"},
190 {IXGBE_RDT(0), "RDT"},
191 {IXGBE_RXDCTL(0), "RXDCTL"},
192 {IXGBE_RDBAL(0), "RDBAL"},
193 {IXGBE_RDBAH(0), "RDBAH"},
196 {IXGBE_TDBAL(0), "TDBAL"},
197 {IXGBE_TDBAH(0), "TDBAH"},
198 {IXGBE_TDLEN(0), "TDLEN"},
199 {IXGBE_TDH(0), "TDH"},
200 {IXGBE_TDT(0), "TDT"},
201 {IXGBE_TXDCTL(0), "TXDCTL"},
203 /* List Terminator */
209 * ixgbe_regdump - register printout routine
211 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
217 switch (reginfo->ofs) {
218 case IXGBE_SRRCTL(0):
219 for (i = 0; i < 64; i++)
220 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
222 case IXGBE_DCA_RXCTRL(0):
223 for (i = 0; i < 64; i++)
224 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
227 for (i = 0; i < 64; i++)
228 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
231 for (i = 0; i < 64; i++)
232 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
235 for (i = 0; i < 64; i++)
236 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
238 case IXGBE_RXDCTL(0):
239 for (i = 0; i < 64; i++)
240 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
243 for (i = 0; i < 64; i++)
244 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
247 for (i = 0; i < 64; i++)
248 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
251 for (i = 0; i < 64; i++)
252 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
255 for (i = 0; i < 64; i++)
256 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
259 for (i = 0; i < 64; i++)
260 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
263 for (i = 0; i < 64; i++)
264 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
267 for (i = 0; i < 64; i++)
268 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
270 case IXGBE_TXDCTL(0):
271 for (i = 0; i < 64; i++)
272 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
275 pr_info("%-15s %08x\n", reginfo->name,
276 IXGBE_READ_REG(hw, reginfo->ofs));
280 for (i = 0; i < 8; i++) {
281 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
282 pr_err("%-15s", rname);
283 for (j = 0; j < 8; j++)
284 pr_cont(" %08x", regs[i*8+j]);
291 * ixgbe_dump - Print registers, tx-rings and rx-rings
293 static void ixgbe_dump(struct ixgbe_adapter *adapter)
295 struct net_device *netdev = adapter->netdev;
296 struct ixgbe_hw *hw = &adapter->hw;
297 struct ixgbe_reg_info *reginfo;
299 struct ixgbe_ring *tx_ring;
300 struct ixgbe_tx_buffer *tx_buffer;
301 union ixgbe_adv_tx_desc *tx_desc;
302 struct my_u0 { u64 a; u64 b; } *u0;
303 struct ixgbe_ring *rx_ring;
304 union ixgbe_adv_rx_desc *rx_desc;
305 struct ixgbe_rx_buffer *rx_buffer_info;
309 if (!netif_msg_hw(adapter))
312 /* Print netdevice Info */
314 dev_info(&adapter->pdev->dev, "Net device Info\n");
315 pr_info("Device Name state "
316 "trans_start last_rx\n");
317 pr_info("%-15s %016lX %016lX %016lX\n",
324 /* Print Registers */
325 dev_info(&adapter->pdev->dev, "Register Dump\n");
326 pr_info(" Register Name Value\n");
327 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
328 reginfo->name; reginfo++) {
329 ixgbe_regdump(hw, reginfo);
332 /* Print TX Ring Summary */
333 if (!netdev || !netif_running(netdev))
336 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
337 pr_info("Queue [NTU] [NTC] [bi(ntc)->dma ] leng ntw timestamp\n");
338 for (n = 0; n < adapter->num_tx_queues; n++) {
339 tx_ring = adapter->tx_ring[n];
340 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
341 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
342 n, tx_ring->next_to_use, tx_ring->next_to_clean,
343 (u64)dma_unmap_addr(tx_buffer, dma),
344 dma_unmap_len(tx_buffer, len),
345 tx_buffer->next_to_watch,
346 (u64)tx_buffer->time_stamp);
350 if (!netif_msg_tx_done(adapter))
351 goto rx_ring_summary;
353 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
355 /* Transmit Descriptor Formats
357 * Advanced Transmit Descriptor
358 * +--------------------------------------------------------------+
359 * 0 | Buffer Address [63:0] |
360 * +--------------------------------------------------------------+
361 * 8 | PAYLEN | PORTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
362 * +--------------------------------------------------------------+
363 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
366 for (n = 0; n < adapter->num_tx_queues; n++) {
367 tx_ring = adapter->tx_ring[n];
368 pr_info("------------------------------------\n");
369 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
370 pr_info("------------------------------------\n");
371 pr_info("T [desc] [address 63:0 ] "
372 "[PlPOIdStDDt Ln] [bi->dma ] "
373 "leng ntw timestamp bi->skb\n");
375 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
376 tx_desc = IXGBE_TX_DESC(tx_ring, i);
377 tx_buffer = &tx_ring->tx_buffer_info[i];
378 u0 = (struct my_u0 *)tx_desc;
379 pr_info("T [0x%03X] %016llX %016llX %016llX"
380 " %04X %p %016llX %p", i,
383 (u64)dma_unmap_addr(tx_buffer, dma),
384 dma_unmap_len(tx_buffer, len),
385 tx_buffer->next_to_watch,
386 (u64)tx_buffer->time_stamp,
388 if (i == tx_ring->next_to_use &&
389 i == tx_ring->next_to_clean)
391 else if (i == tx_ring->next_to_use)
393 else if (i == tx_ring->next_to_clean)
398 if (netif_msg_pktdata(adapter) &&
399 dma_unmap_len(tx_buffer, len) != 0)
400 print_hex_dump(KERN_INFO, "",
401 DUMP_PREFIX_ADDRESS, 16, 1,
402 phys_to_virt(dma_unmap_addr(tx_buffer,
404 dma_unmap_len(tx_buffer, len),
409 /* Print RX Rings Summary */
411 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
412 pr_info("Queue [NTU] [NTC]\n");
413 for (n = 0; n < adapter->num_rx_queues; n++) {
414 rx_ring = adapter->rx_ring[n];
415 pr_info("%5d %5X %5X\n",
416 n, rx_ring->next_to_use, rx_ring->next_to_clean);
420 if (!netif_msg_rx_status(adapter))
423 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
425 /* Advanced Receive Descriptor (Read) Format
427 * +-----------------------------------------------------+
428 * 0 | Packet Buffer Address [63:1] |A0/NSE|
429 * +----------------------------------------------+------+
430 * 8 | Header Buffer Address [63:1] | DD |
431 * +-----------------------------------------------------+
434 * Advanced Receive Descriptor (Write-Back) Format
436 * 63 48 47 32 31 30 21 20 16 15 4 3 0
437 * +------------------------------------------------------+
438 * 0 | Packet IP |SPH| HDR_LEN | RSV|Packet| RSS |
439 * | Checksum Ident | | | | Type | Type |
440 * +------------------------------------------------------+
441 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
442 * +------------------------------------------------------+
443 * 63 48 47 32 31 20 19 0
445 for (n = 0; n < adapter->num_rx_queues; n++) {
446 rx_ring = adapter->rx_ring[n];
447 pr_info("------------------------------------\n");
448 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
449 pr_info("------------------------------------\n");
450 pr_info("R [desc] [ PktBuf A0] "
451 "[ HeadBuf DD] [bi->dma ] [bi->skb] "
452 "<-- Adv Rx Read format\n");
453 pr_info("RWB[desc] [PcsmIpSHl PtRs] "
454 "[vl er S cks ln] ---------------- [bi->skb] "
455 "<-- Adv Rx Write-Back format\n");
457 for (i = 0; i < rx_ring->count; i++) {
458 rx_buffer_info = &rx_ring->rx_buffer_info[i];
459 rx_desc = IXGBE_RX_DESC(rx_ring, i);
460 u0 = (struct my_u0 *)rx_desc;
461 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
462 if (staterr & IXGBE_RXD_STAT_DD) {
463 /* Descriptor Done */
464 pr_info("RWB[0x%03X] %016llX "
465 "%016llX ---------------- %p", i,
468 rx_buffer_info->skb);
470 pr_info("R [0x%03X] %016llX "
471 "%016llX %016llX %p", i,
474 (u64)rx_buffer_info->dma,
475 rx_buffer_info->skb);
477 if (netif_msg_pktdata(adapter)) {
478 print_hex_dump(KERN_INFO, "",
479 DUMP_PREFIX_ADDRESS, 16, 1,
480 phys_to_virt(rx_buffer_info->dma),
481 ixgbe_rx_bufsz(rx_ring), true);
485 if (i == rx_ring->next_to_use)
487 else if (i == rx_ring->next_to_clean)
499 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
503 /* Let firmware take over control of h/w */
504 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
505 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
506 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
509 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
513 /* Let firmware know the driver has taken over */
514 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
515 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
516 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
520 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
521 * @adapter: pointer to adapter struct
522 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
523 * @queue: queue to map the corresponding interrupt to
524 * @msix_vector: the vector to map to the corresponding queue
527 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
528 u8 queue, u8 msix_vector)
531 struct ixgbe_hw *hw = &adapter->hw;
532 switch (hw->mac.type) {
533 case ixgbe_mac_82598EB:
534 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
537 index = (((direction * 64) + queue) >> 2) & 0x1F;
538 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
539 ivar &= ~(0xFF << (8 * (queue & 0x3)));
540 ivar |= (msix_vector << (8 * (queue & 0x3)));
541 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
543 case ixgbe_mac_82599EB:
545 if (direction == -1) {
547 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
548 index = ((queue & 1) * 8);
549 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
550 ivar &= ~(0xFF << index);
551 ivar |= (msix_vector << index);
552 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
555 /* tx or rx causes */
556 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
557 index = ((16 * (queue & 1)) + (8 * direction));
558 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
559 ivar &= ~(0xFF << index);
560 ivar |= (msix_vector << index);
561 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
569 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
574 switch (adapter->hw.mac.type) {
575 case ixgbe_mac_82598EB:
576 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
577 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
579 case ixgbe_mac_82599EB:
581 mask = (qmask & 0xFFFFFFFF);
582 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
583 mask = (qmask >> 32);
584 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
591 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
592 struct ixgbe_tx_buffer *tx_buffer)
594 if (tx_buffer->skb) {
595 dev_kfree_skb_any(tx_buffer->skb);
596 if (dma_unmap_len(tx_buffer, len))
597 dma_unmap_single(ring->dev,
598 dma_unmap_addr(tx_buffer, dma),
599 dma_unmap_len(tx_buffer, len),
601 } else if (dma_unmap_len(tx_buffer, len)) {
602 dma_unmap_page(ring->dev,
603 dma_unmap_addr(tx_buffer, dma),
604 dma_unmap_len(tx_buffer, len),
607 tx_buffer->next_to_watch = NULL;
608 tx_buffer->skb = NULL;
609 dma_unmap_len_set(tx_buffer, len, 0);
610 /* tx_buffer must be completely set up in the transmit path */
613 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
615 struct ixgbe_hw *hw = &adapter->hw;
616 struct ixgbe_hw_stats *hwstats = &adapter->stats;
620 if ((hw->fc.current_mode != ixgbe_fc_full) &&
621 (hw->fc.current_mode != ixgbe_fc_rx_pause))
624 switch (hw->mac.type) {
625 case ixgbe_mac_82598EB:
626 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
629 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
631 hwstats->lxoffrxc += data;
633 /* refill credits (no tx hang) if we received xoff */
637 for (i = 0; i < adapter->num_tx_queues; i++)
638 clear_bit(__IXGBE_HANG_CHECK_ARMED,
639 &adapter->tx_ring[i]->state);
642 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
644 struct ixgbe_hw *hw = &adapter->hw;
645 struct ixgbe_hw_stats *hwstats = &adapter->stats;
648 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
650 if (adapter->ixgbe_ieee_pfc)
651 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
653 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
654 ixgbe_update_xoff_rx_lfc(adapter);
658 /* update stats for each tc, only valid with PFC enabled */
659 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
660 switch (hw->mac.type) {
661 case ixgbe_mac_82598EB:
662 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
665 xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
667 hwstats->pxoffrxc[i] += xoff[i];
670 /* disarm tx queues that have received xoff frames */
671 for (i = 0; i < adapter->num_tx_queues; i++) {
672 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
673 u8 tc = tx_ring->dcb_tc;
676 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
680 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
682 return ring->stats.packets;
685 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
687 struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
688 struct ixgbe_hw *hw = &adapter->hw;
690 u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
691 u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
694 return (head < tail) ?
695 tail - head : (tail + ring->count - head);
700 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
702 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
703 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
704 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
707 clear_check_for_tx_hang(tx_ring);
710 * Check for a hung queue, but be thorough. This verifies
711 * that a transmit has been completed since the previous
712 * check AND there is at least one packet pending. The
713 * ARMED bit is set to indicate a potential hang. The
714 * bit is cleared if a pause frame is received to remove
715 * false hang detection due to PFC or 802.3x frames. By
716 * requiring this to fail twice we avoid races with
717 * pfc clearing the ARMED bit and conditions where we
718 * run the check_tx_hang logic with a transmit completion
719 * pending but without time to complete it yet.
721 if ((tx_done_old == tx_done) && tx_pending) {
722 /* make sure it is true for two checks in a row */
723 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
726 /* update completed stats and continue */
727 tx_ring->tx_stats.tx_done_old = tx_done;
728 /* reset the countdown */
729 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
736 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
737 * @adapter: driver private struct
739 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
742 /* Do the reset outside of interrupt context */
743 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
744 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
745 ixgbe_service_event_schedule(adapter);
750 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
751 * @q_vector: structure containing interrupt and ring information
752 * @tx_ring: tx ring to clean
754 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
755 struct ixgbe_ring *tx_ring)
757 struct ixgbe_adapter *adapter = q_vector->adapter;
758 struct ixgbe_tx_buffer *tx_buffer;
759 union ixgbe_adv_tx_desc *tx_desc;
760 unsigned int total_bytes = 0, total_packets = 0;
761 unsigned int budget = q_vector->tx.work_limit;
762 unsigned int i = tx_ring->next_to_clean;
764 if (test_bit(__IXGBE_DOWN, &adapter->state))
767 tx_buffer = &tx_ring->tx_buffer_info[i];
768 tx_desc = IXGBE_TX_DESC(tx_ring, i);
772 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
774 /* if next_to_watch is not set then there is no work pending */
778 /* prevent any other reads prior to eop_desc */
781 /* if DD is not set pending work has not been completed */
782 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
785 /* clear next_to_watch to prevent false hangs */
786 tx_buffer->next_to_watch = NULL;
788 /* update the statistics for this packet */
789 total_bytes += tx_buffer->bytecount;
790 total_packets += tx_buffer->gso_segs;
792 #ifdef CONFIG_IXGBE_PTP
793 if (unlikely(tx_buffer->tx_flags & IXGBE_TX_FLAGS_TSTAMP))
794 ixgbe_ptp_tx_hwtstamp(q_vector, tx_buffer->skb);
798 dev_kfree_skb_any(tx_buffer->skb);
800 /* unmap skb header data */
801 dma_unmap_single(tx_ring->dev,
802 dma_unmap_addr(tx_buffer, dma),
803 dma_unmap_len(tx_buffer, len),
806 /* clear tx_buffer data */
807 tx_buffer->skb = NULL;
808 dma_unmap_len_set(tx_buffer, len, 0);
810 /* unmap remaining buffers */
811 while (tx_desc != eop_desc) {
817 tx_buffer = tx_ring->tx_buffer_info;
818 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
821 /* unmap any remaining paged data */
822 if (dma_unmap_len(tx_buffer, len)) {
823 dma_unmap_page(tx_ring->dev,
824 dma_unmap_addr(tx_buffer, dma),
825 dma_unmap_len(tx_buffer, len),
827 dma_unmap_len_set(tx_buffer, len, 0);
831 /* move us one more past the eop_desc for start of next pkt */
837 tx_buffer = tx_ring->tx_buffer_info;
838 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
841 /* issue prefetch for next Tx descriptor */
844 /* update budget accounting */
846 } while (likely(budget));
849 tx_ring->next_to_clean = i;
850 u64_stats_update_begin(&tx_ring->syncp);
851 tx_ring->stats.bytes += total_bytes;
852 tx_ring->stats.packets += total_packets;
853 u64_stats_update_end(&tx_ring->syncp);
854 q_vector->tx.total_bytes += total_bytes;
855 q_vector->tx.total_packets += total_packets;
857 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
858 /* schedule immediate reset if we believe we hung */
859 struct ixgbe_hw *hw = &adapter->hw;
860 e_err(drv, "Detected Tx Unit Hang\n"
862 " TDH, TDT <%x>, <%x>\n"
863 " next_to_use <%x>\n"
864 " next_to_clean <%x>\n"
865 "tx_buffer_info[next_to_clean]\n"
866 " time_stamp <%lx>\n"
868 tx_ring->queue_index,
869 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
870 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
871 tx_ring->next_to_use, i,
872 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
874 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
877 "tx hang %d detected on queue %d, resetting adapter\n",
878 adapter->tx_timeout_count + 1, tx_ring->queue_index);
880 /* schedule immediate reset if we believe we hung */
881 ixgbe_tx_timeout_reset(adapter);
883 /* the adapter is about to reset, no point in enabling stuff */
887 netdev_tx_completed_queue(txring_txq(tx_ring),
888 total_packets, total_bytes);
890 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
891 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
892 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
893 /* Make sure that anybody stopping the queue after this
894 * sees the new next_to_clean.
897 if (__netif_subqueue_stopped(tx_ring->netdev,
898 tx_ring->queue_index)
899 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
900 netif_wake_subqueue(tx_ring->netdev,
901 tx_ring->queue_index);
902 ++tx_ring->tx_stats.restart_queue;
909 #ifdef CONFIG_IXGBE_DCA
910 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
911 struct ixgbe_ring *tx_ring,
914 struct ixgbe_hw *hw = &adapter->hw;
915 u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
918 switch (hw->mac.type) {
919 case ixgbe_mac_82598EB:
920 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
922 case ixgbe_mac_82599EB:
924 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
925 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
928 /* for unknown hardware do not write register */
933 * We can enable relaxed ordering for reads, but not writes when
934 * DCA is enabled. This is due to a known issue in some chipsets
935 * which will cause the DCA tag to be cleared.
937 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
938 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
939 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
941 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
944 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
945 struct ixgbe_ring *rx_ring,
948 struct ixgbe_hw *hw = &adapter->hw;
949 u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
950 u8 reg_idx = rx_ring->reg_idx;
953 switch (hw->mac.type) {
954 case ixgbe_mac_82599EB:
956 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
963 * We can enable relaxed ordering for reads, but not writes when
964 * DCA is enabled. This is due to a known issue in some chipsets
965 * which will cause the DCA tag to be cleared.
967 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
968 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
969 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
971 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
974 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
976 struct ixgbe_adapter *adapter = q_vector->adapter;
977 struct ixgbe_ring *ring;
980 if (q_vector->cpu == cpu)
983 ixgbe_for_each_ring(ring, q_vector->tx)
984 ixgbe_update_tx_dca(adapter, ring, cpu);
986 ixgbe_for_each_ring(ring, q_vector->rx)
987 ixgbe_update_rx_dca(adapter, ring, cpu);
994 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
998 if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1001 /* always use CB2 mode, difference is masked in the CB driver */
1002 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1004 for (i = 0; i < adapter->num_q_vectors; i++) {
1005 adapter->q_vector[i]->cpu = -1;
1006 ixgbe_update_dca(adapter->q_vector[i]);
1010 static int __ixgbe_notify_dca(struct device *dev, void *data)
1012 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1013 unsigned long event = *(unsigned long *)data;
1015 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1019 case DCA_PROVIDER_ADD:
1020 /* if we're already enabled, don't do it again */
1021 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1023 if (dca_add_requester(dev) == 0) {
1024 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1025 ixgbe_setup_dca(adapter);
1028 /* Fall Through since DCA is disabled. */
1029 case DCA_PROVIDER_REMOVE:
1030 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1031 dca_remove_requester(dev);
1032 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1033 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1041 #endif /* CONFIG_IXGBE_DCA */
1042 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1043 union ixgbe_adv_rx_desc *rx_desc,
1044 struct sk_buff *skb)
1046 if (ring->netdev->features & NETIF_F_RXHASH)
1047 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1052 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1053 * @ring: structure containing ring specific data
1054 * @rx_desc: advanced rx descriptor
1056 * Returns : true if it is FCoE pkt
1058 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1059 union ixgbe_adv_rx_desc *rx_desc)
1061 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1063 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1064 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1065 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1066 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1069 #endif /* IXGBE_FCOE */
1071 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1072 * @ring: structure containing ring specific data
1073 * @rx_desc: current Rx descriptor being processed
1074 * @skb: skb currently being received and modified
1076 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1077 union ixgbe_adv_rx_desc *rx_desc,
1078 struct sk_buff *skb)
1080 skb_checksum_none_assert(skb);
1082 /* Rx csum disabled */
1083 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1086 /* if IP and error */
1087 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1088 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1089 ring->rx_stats.csum_err++;
1093 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1096 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1097 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1100 * 82599 errata, UDP frames with a 0 checksum can be marked as
1103 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1104 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1107 ring->rx_stats.csum_err++;
1111 /* It must be a TCP or UDP packet with a valid checksum */
1112 skb->ip_summed = CHECKSUM_UNNECESSARY;
1115 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1117 rx_ring->next_to_use = val;
1119 /* update next to alloc since we have filled the ring */
1120 rx_ring->next_to_alloc = val;
1122 * Force memory writes to complete before letting h/w
1123 * know there are new descriptors to fetch. (Only
1124 * applicable for weak-ordered memory model archs,
1128 writel(val, rx_ring->tail);
1131 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1132 struct ixgbe_rx_buffer *bi)
1134 struct page *page = bi->page;
1135 dma_addr_t dma = bi->dma;
1137 /* since we are recycling buffers we should seldom need to alloc */
1141 /* alloc new page for storage */
1142 if (likely(!page)) {
1143 page = alloc_pages(GFP_ATOMIC | __GFP_COLD | __GFP_COMP,
1144 ixgbe_rx_pg_order(rx_ring));
1145 if (unlikely(!page)) {
1146 rx_ring->rx_stats.alloc_rx_page_failed++;
1152 /* map page for use */
1153 dma = dma_map_page(rx_ring->dev, page, 0,
1154 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1157 * if mapping failed free memory back to system since
1158 * there isn't much point in holding memory we can't use
1160 if (dma_mapping_error(rx_ring->dev, dma)) {
1161 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1164 rx_ring->rx_stats.alloc_rx_page_failed++;
1169 bi->page_offset ^= ixgbe_rx_bufsz(rx_ring);
1175 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1176 * @rx_ring: ring to place buffers on
1177 * @cleaned_count: number of buffers to replace
1179 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1181 union ixgbe_adv_rx_desc *rx_desc;
1182 struct ixgbe_rx_buffer *bi;
1183 u16 i = rx_ring->next_to_use;
1189 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1190 bi = &rx_ring->rx_buffer_info[i];
1191 i -= rx_ring->count;
1194 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1198 * Refresh the desc even if buffer_addrs didn't change
1199 * because each write-back erases this info.
1201 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1207 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1208 bi = rx_ring->rx_buffer_info;
1209 i -= rx_ring->count;
1212 /* clear the hdr_addr for the next_to_use descriptor */
1213 rx_desc->read.hdr_addr = 0;
1216 } while (cleaned_count);
1218 i += rx_ring->count;
1220 if (rx_ring->next_to_use != i)
1221 ixgbe_release_rx_desc(rx_ring, i);
1225 * ixgbe_get_headlen - determine size of header for RSC/LRO/GRO/FCOE
1226 * @data: pointer to the start of the headers
1227 * @max_len: total length of section to find headers in
1229 * This function is meant to determine the length of headers that will
1230 * be recognized by hardware for LRO, GRO, and RSC offloads. The main
1231 * motivation of doing this is to only perform one pull for IPv4 TCP
1232 * packets so that we can do basic things like calculating the gso_size
1233 * based on the average data per packet.
1235 static unsigned int ixgbe_get_headlen(unsigned char *data,
1236 unsigned int max_len)
1239 unsigned char *network;
1242 struct vlan_hdr *vlan;
1247 u8 nexthdr = 0; /* default to not TCP */
1250 /* this should never happen, but better safe than sorry */
1251 if (max_len < ETH_HLEN)
1254 /* initialize network frame pointer */
1257 /* set first protocol and move network header forward */
1258 protocol = hdr.eth->h_proto;
1259 hdr.network += ETH_HLEN;
1261 /* handle any vlan tag if present */
1262 if (protocol == __constant_htons(ETH_P_8021Q)) {
1263 if ((hdr.network - data) > (max_len - VLAN_HLEN))
1266 protocol = hdr.vlan->h_vlan_encapsulated_proto;
1267 hdr.network += VLAN_HLEN;
1270 /* handle L3 protocols */
1271 if (protocol == __constant_htons(ETH_P_IP)) {
1272 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
1275 /* access ihl as a u8 to avoid unaligned access on ia64 */
1276 hlen = (hdr.network[0] & 0x0F) << 2;
1278 /* verify hlen meets minimum size requirements */
1279 if (hlen < sizeof(struct iphdr))
1280 return hdr.network - data;
1282 /* record next protocol */
1283 nexthdr = hdr.ipv4->protocol;
1284 hdr.network += hlen;
1286 } else if (protocol == __constant_htons(ETH_P_FCOE)) {
1287 if ((hdr.network - data) > (max_len - FCOE_HEADER_LEN))
1289 hdr.network += FCOE_HEADER_LEN;
1292 return hdr.network - data;
1295 /* finally sort out TCP */
1296 if (nexthdr == IPPROTO_TCP) {
1297 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
1300 /* access doff as a u8 to avoid unaligned access on ia64 */
1301 hlen = (hdr.network[12] & 0xF0) >> 2;
1303 /* verify hlen meets minimum size requirements */
1304 if (hlen < sizeof(struct tcphdr))
1305 return hdr.network - data;
1307 hdr.network += hlen;
1311 * If everything has gone correctly hdr.network should be the
1312 * data section of the packet and will be the end of the header.
1313 * If not then it probably represents the end of the last recognized
1316 if ((hdr.network - data) < max_len)
1317 return hdr.network - data;
1322 static void ixgbe_get_rsc_cnt(struct ixgbe_ring *rx_ring,
1323 union ixgbe_adv_rx_desc *rx_desc,
1324 struct sk_buff *skb)
1329 if (!ring_is_rsc_enabled(rx_ring))
1332 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1333 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1335 /* If this is an RSC frame rsc_cnt should be non-zero */
1339 rsc_cnt = le32_to_cpu(rsc_enabled);
1340 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1342 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1345 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1346 struct sk_buff *skb)
1348 u16 hdr_len = skb_headlen(skb);
1350 /* set gso_size to avoid messing up TCP MSS */
1351 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1352 IXGBE_CB(skb)->append_cnt);
1355 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1356 struct sk_buff *skb)
1358 /* if append_cnt is 0 then frame is not RSC */
1359 if (!IXGBE_CB(skb)->append_cnt)
1362 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1363 rx_ring->rx_stats.rsc_flush++;
1365 ixgbe_set_rsc_gso_size(rx_ring, skb);
1367 /* gso_size is computed using append_cnt so always clear it last */
1368 IXGBE_CB(skb)->append_cnt = 0;
1372 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1373 * @rx_ring: rx descriptor ring packet is being transacted on
1374 * @rx_desc: pointer to the EOP Rx descriptor
1375 * @skb: pointer to current skb being populated
1377 * This function checks the ring, descriptor, and packet information in
1378 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1379 * other fields within the skb.
1381 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1382 union ixgbe_adv_rx_desc *rx_desc,
1383 struct sk_buff *skb)
1385 struct net_device *dev = rx_ring->netdev;
1387 ixgbe_update_rsc_stats(rx_ring, skb);
1389 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1391 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1393 #ifdef CONFIG_IXGBE_PTP
1394 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
1397 if ((dev->features & NETIF_F_HW_VLAN_RX) &&
1398 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1399 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1400 __vlan_hwaccel_put_tag(skb, vid);
1403 skb_record_rx_queue(skb, rx_ring->queue_index);
1405 skb->protocol = eth_type_trans(skb, dev);
1408 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1409 struct sk_buff *skb)
1411 struct ixgbe_adapter *adapter = q_vector->adapter;
1413 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1414 napi_gro_receive(&q_vector->napi, skb);
1420 * ixgbe_is_non_eop - process handling of non-EOP buffers
1421 * @rx_ring: Rx ring being processed
1422 * @rx_desc: Rx descriptor for current buffer
1423 * @skb: Current socket buffer containing buffer in progress
1425 * This function updates next to clean. If the buffer is an EOP buffer
1426 * this function exits returning false, otherwise it will place the
1427 * sk_buff in the next buffer to be chained and return true indicating
1428 * that this is in fact a non-EOP buffer.
1430 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1431 union ixgbe_adv_rx_desc *rx_desc,
1432 struct sk_buff *skb)
1434 u32 ntc = rx_ring->next_to_clean + 1;
1436 /* fetch, update, and store next to clean */
1437 ntc = (ntc < rx_ring->count) ? ntc : 0;
1438 rx_ring->next_to_clean = ntc;
1440 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1442 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1445 /* append_cnt indicates packet is RSC, if so fetch nextp */
1446 if (IXGBE_CB(skb)->append_cnt) {
1447 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1448 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1449 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1452 /* place skb in next buffer to be received */
1453 rx_ring->rx_buffer_info[ntc].skb = skb;
1454 rx_ring->rx_stats.non_eop_descs++;
1460 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1461 * @rx_ring: rx descriptor ring packet is being transacted on
1462 * @rx_desc: pointer to the EOP Rx descriptor
1463 * @skb: pointer to current skb being fixed
1465 * Check for corrupted packet headers caused by senders on the local L2
1466 * embedded NIC switch not setting up their Tx Descriptors right. These
1467 * should be very rare.
1469 * Also address the case where we are pulling data in on pages only
1470 * and as such no data is present in the skb header.
1472 * In addition if skb is not at least 60 bytes we need to pad it so that
1473 * it is large enough to qualify as a valid Ethernet frame.
1475 * Returns true if an error was encountered and skb was freed.
1477 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1478 union ixgbe_adv_rx_desc *rx_desc,
1479 struct sk_buff *skb)
1481 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1482 struct net_device *netdev = rx_ring->netdev;
1484 unsigned int pull_len;
1486 /* if the page was released unmap it, else just sync our portion */
1487 if (unlikely(IXGBE_CB(skb)->page_released)) {
1488 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1489 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1490 IXGBE_CB(skb)->page_released = false;
1492 dma_sync_single_range_for_cpu(rx_ring->dev,
1495 ixgbe_rx_bufsz(rx_ring),
1498 IXGBE_CB(skb)->dma = 0;
1500 /* verify that the packet does not have any known errors */
1501 if (unlikely(ixgbe_test_staterr(rx_desc,
1502 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1503 !(netdev->features & NETIF_F_RXALL))) {
1504 dev_kfree_skb_any(skb);
1509 * it is valid to use page_address instead of kmap since we are
1510 * working with pages allocated out of the lomem pool per
1511 * alloc_page(GFP_ATOMIC)
1513 va = skb_frag_address(frag);
1516 * we need the header to contain the greater of either ETH_HLEN or
1517 * 60 bytes if the skb->len is less than 60 for skb_pad.
1519 pull_len = skb_frag_size(frag);
1521 pull_len = ixgbe_get_headlen(va, pull_len);
1523 /* align pull length to size of long to optimize memcpy performance */
1524 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1526 /* update all of the pointers */
1527 skb_frag_size_sub(frag, pull_len);
1528 frag->page_offset += pull_len;
1529 skb->data_len -= pull_len;
1530 skb->tail += pull_len;
1533 * if we sucked the frag empty then we should free it,
1534 * if there are other frags here something is screwed up in hardware
1536 if (skb_frag_size(frag) == 0) {
1537 BUG_ON(skb_shinfo(skb)->nr_frags != 1);
1538 skb_shinfo(skb)->nr_frags = 0;
1539 __skb_frag_unref(frag);
1540 skb->truesize -= ixgbe_rx_bufsz(rx_ring);
1544 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1545 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1549 /* if skb_pad returns an error the skb was freed */
1550 if (unlikely(skb->len < 60)) {
1551 int pad_len = 60 - skb->len;
1553 if (skb_pad(skb, pad_len))
1555 __skb_put(skb, pad_len);
1562 * ixgbe_can_reuse_page - determine if we can reuse a page
1563 * @rx_buffer: pointer to rx_buffer containing the page we want to reuse
1565 * Returns true if page can be reused in another Rx buffer
1567 static inline bool ixgbe_can_reuse_page(struct ixgbe_rx_buffer *rx_buffer)
1569 struct page *page = rx_buffer->page;
1571 /* if we are only owner of page and it is local we can reuse it */
1572 return likely(page_count(page) == 1) &&
1573 likely(page_to_nid(page) == numa_node_id());
1577 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1578 * @rx_ring: rx descriptor ring to store buffers on
1579 * @old_buff: donor buffer to have page reused
1581 * Syncronizes page for reuse by the adapter
1583 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1584 struct ixgbe_rx_buffer *old_buff)
1586 struct ixgbe_rx_buffer *new_buff;
1587 u16 nta = rx_ring->next_to_alloc;
1588 u16 bufsz = ixgbe_rx_bufsz(rx_ring);
1590 new_buff = &rx_ring->rx_buffer_info[nta];
1592 /* update, and store next to alloc */
1594 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1596 /* transfer page from old buffer to new buffer */
1597 new_buff->page = old_buff->page;
1598 new_buff->dma = old_buff->dma;
1600 /* flip page offset to other buffer and store to new_buff */
1601 new_buff->page_offset = old_buff->page_offset ^ bufsz;
1603 /* sync the buffer for use by the device */
1604 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1605 new_buff->page_offset, bufsz,
1608 /* bump ref count on page before it is given to the stack */
1609 get_page(new_buff->page);
1613 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1614 * @rx_ring: rx descriptor ring to transact packets on
1615 * @rx_buffer: buffer containing page to add
1616 * @rx_desc: descriptor containing length of buffer written by hardware
1617 * @skb: sk_buff to place the data into
1619 * This function is based on skb_add_rx_frag. I would have used that
1620 * function however it doesn't handle the truesize case correctly since we
1621 * are allocating more memory than might be used for a single receive.
1623 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1624 struct ixgbe_rx_buffer *rx_buffer,
1625 struct sk_buff *skb, int size)
1627 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1628 rx_buffer->page, rx_buffer->page_offset,
1631 skb->data_len += size;
1632 skb->truesize += ixgbe_rx_bufsz(rx_ring);
1636 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1637 * @q_vector: structure containing interrupt and ring information
1638 * @rx_ring: rx descriptor ring to transact packets on
1639 * @budget: Total limit on number of packets to process
1641 * This function provides a "bounce buffer" approach to Rx interrupt
1642 * processing. The advantage to this is that on systems that have
1643 * expensive overhead for IOMMU access this provides a means of avoiding
1644 * it by maintaining the mapping of the page to the syste.
1646 * Returns true if all work is completed without reaching budget
1648 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1649 struct ixgbe_ring *rx_ring,
1652 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1654 struct ixgbe_adapter *adapter = q_vector->adapter;
1656 #endif /* IXGBE_FCOE */
1657 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
1660 struct ixgbe_rx_buffer *rx_buffer;
1661 union ixgbe_adv_rx_desc *rx_desc;
1662 struct sk_buff *skb;
1666 /* return some buffers to hardware, one at a time is too slow */
1667 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1668 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1672 ntc = rx_ring->next_to_clean;
1673 rx_desc = IXGBE_RX_DESC(rx_ring, ntc);
1674 rx_buffer = &rx_ring->rx_buffer_info[ntc];
1676 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
1680 * This memory barrier is needed to keep us from reading
1681 * any other fields out of the rx_desc until we know the
1682 * RXD_STAT_DD bit is set
1686 page = rx_buffer->page;
1689 skb = rx_buffer->skb;
1692 void *page_addr = page_address(page) +
1693 rx_buffer->page_offset;
1695 /* prefetch first cache line of first page */
1696 prefetch(page_addr);
1697 #if L1_CACHE_BYTES < 128
1698 prefetch(page_addr + L1_CACHE_BYTES);
1701 /* allocate a skb to store the frags */
1702 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1704 if (unlikely(!skb)) {
1705 rx_ring->rx_stats.alloc_rx_buff_failed++;
1710 * we will be copying header into skb->data in
1711 * pskb_may_pull so it is in our interest to prefetch
1712 * it now to avoid a possible cache miss
1714 prefetchw(skb->data);
1717 * Delay unmapping of the first packet. It carries the
1718 * header information, HW may still access the header
1719 * after the writeback. Only unmap it when EOP is
1722 IXGBE_CB(skb)->dma = rx_buffer->dma;
1724 /* we are reusing so sync this buffer for CPU use */
1725 dma_sync_single_range_for_cpu(rx_ring->dev,
1727 rx_buffer->page_offset,
1728 ixgbe_rx_bufsz(rx_ring),
1732 /* pull page into skb */
1733 ixgbe_add_rx_frag(rx_ring, rx_buffer, skb,
1734 le16_to_cpu(rx_desc->wb.upper.length));
1736 if (ixgbe_can_reuse_page(rx_buffer)) {
1737 /* hand second half of page back to the ring */
1738 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
1739 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
1740 /* the page has been released from the ring */
1741 IXGBE_CB(skb)->page_released = true;
1743 /* we are not reusing the buffer so unmap it */
1744 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
1745 ixgbe_rx_pg_size(rx_ring),
1749 /* clear contents of buffer_info */
1750 rx_buffer->skb = NULL;
1752 rx_buffer->page = NULL;
1754 ixgbe_get_rsc_cnt(rx_ring, rx_desc, skb);
1758 /* place incomplete frames back on ring for completion */
1759 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
1762 /* verify the packet layout is correct */
1763 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
1766 /* probably a little skewed due to removing CRC */
1767 total_rx_bytes += skb->len;
1770 /* populate checksum, timestamp, VLAN, and protocol */
1771 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
1774 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1775 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
1776 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1778 dev_kfree_skb_any(skb);
1783 #endif /* IXGBE_FCOE */
1784 ixgbe_rx_skb(q_vector, skb);
1786 /* update budget accounting */
1788 } while (likely(budget));
1791 /* include DDPed FCoE data */
1792 if (ddp_bytes > 0) {
1795 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1796 sizeof(struct fc_frame_header) -
1797 sizeof(struct fcoe_crc_eof);
1800 total_rx_bytes += ddp_bytes;
1801 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1804 #endif /* IXGBE_FCOE */
1805 u64_stats_update_begin(&rx_ring->syncp);
1806 rx_ring->stats.packets += total_rx_packets;
1807 rx_ring->stats.bytes += total_rx_bytes;
1808 u64_stats_update_end(&rx_ring->syncp);
1809 q_vector->rx.total_packets += total_rx_packets;
1810 q_vector->rx.total_bytes += total_rx_bytes;
1813 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1819 * ixgbe_configure_msix - Configure MSI-X hardware
1820 * @adapter: board private structure
1822 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1825 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1827 struct ixgbe_q_vector *q_vector;
1831 /* Populate MSIX to EITR Select */
1832 if (adapter->num_vfs > 32) {
1833 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
1834 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
1838 * Populate the IVAR table and set the ITR values to the
1839 * corresponding register.
1841 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
1842 struct ixgbe_ring *ring;
1843 q_vector = adapter->q_vector[v_idx];
1845 ixgbe_for_each_ring(ring, q_vector->rx)
1846 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1848 ixgbe_for_each_ring(ring, q_vector->tx)
1849 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1851 if (q_vector->tx.ring && !q_vector->rx.ring) {
1852 /* tx only vector */
1853 if (adapter->tx_itr_setting == 1)
1854 q_vector->itr = IXGBE_10K_ITR;
1856 q_vector->itr = adapter->tx_itr_setting;
1858 /* rx or rx/tx vector */
1859 if (adapter->rx_itr_setting == 1)
1860 q_vector->itr = IXGBE_20K_ITR;
1862 q_vector->itr = adapter->rx_itr_setting;
1865 ixgbe_write_eitr(q_vector);
1868 switch (adapter->hw.mac.type) {
1869 case ixgbe_mac_82598EB:
1870 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1873 case ixgbe_mac_82599EB:
1874 case ixgbe_mac_X540:
1875 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1880 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1882 /* set up to autoclear timer, and the vectors */
1883 mask = IXGBE_EIMS_ENABLE_MASK;
1884 mask &= ~(IXGBE_EIMS_OTHER |
1885 IXGBE_EIMS_MAILBOX |
1888 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1891 enum latency_range {
1895 latency_invalid = 255
1899 * ixgbe_update_itr - update the dynamic ITR value based on statistics
1900 * @q_vector: structure containing interrupt and ring information
1901 * @ring_container: structure containing ring performance data
1903 * Stores a new ITR value based on packets and byte
1904 * counts during the last interrupt. The advantage of per interrupt
1905 * computation is faster updates and more accurate ITR for the current
1906 * traffic pattern. Constants in this function were computed
1907 * based on theoretical maximum wire speed and thresholds were set based
1908 * on testing data as well as attempting to minimize response time
1909 * while increasing bulk throughput.
1910 * this functionality is controlled by the InterruptThrottleRate module
1911 * parameter (see ixgbe_param.c)
1913 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
1914 struct ixgbe_ring_container *ring_container)
1916 int bytes = ring_container->total_bytes;
1917 int packets = ring_container->total_packets;
1920 u8 itr_setting = ring_container->itr;
1925 /* simple throttlerate management
1926 * 0-10MB/s lowest (100000 ints/s)
1927 * 10-20MB/s low (20000 ints/s)
1928 * 20-1249MB/s bulk (8000 ints/s)
1930 /* what was last interrupt timeslice? */
1931 timepassed_us = q_vector->itr >> 2;
1932 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1934 switch (itr_setting) {
1935 case lowest_latency:
1936 if (bytes_perint > 10)
1937 itr_setting = low_latency;
1940 if (bytes_perint > 20)
1941 itr_setting = bulk_latency;
1942 else if (bytes_perint <= 10)
1943 itr_setting = lowest_latency;
1946 if (bytes_perint <= 20)
1947 itr_setting = low_latency;
1951 /* clear work counters since we have the values we need */
1952 ring_container->total_bytes = 0;
1953 ring_container->total_packets = 0;
1955 /* write updated itr to ring container */
1956 ring_container->itr = itr_setting;
1960 * ixgbe_write_eitr - write EITR register in hardware specific way
1961 * @q_vector: structure containing interrupt and ring information
1963 * This function is made to be called by ethtool and by the driver
1964 * when it needs to update EITR registers at runtime. Hardware
1965 * specific quirks/differences are taken care of here.
1967 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1969 struct ixgbe_adapter *adapter = q_vector->adapter;
1970 struct ixgbe_hw *hw = &adapter->hw;
1971 int v_idx = q_vector->v_idx;
1972 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1974 switch (adapter->hw.mac.type) {
1975 case ixgbe_mac_82598EB:
1976 /* must write high and low 16 bits to reset counter */
1977 itr_reg |= (itr_reg << 16);
1979 case ixgbe_mac_82599EB:
1980 case ixgbe_mac_X540:
1982 * set the WDIS bit to not clear the timer bits and cause an
1983 * immediate assertion of the interrupt
1985 itr_reg |= IXGBE_EITR_CNT_WDIS;
1990 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1993 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
1995 u32 new_itr = q_vector->itr;
1998 ixgbe_update_itr(q_vector, &q_vector->tx);
1999 ixgbe_update_itr(q_vector, &q_vector->rx);
2001 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2003 switch (current_itr) {
2004 /* counts and packets in update_itr are dependent on these numbers */
2005 case lowest_latency:
2006 new_itr = IXGBE_100K_ITR;
2009 new_itr = IXGBE_20K_ITR;
2012 new_itr = IXGBE_8K_ITR;
2018 if (new_itr != q_vector->itr) {
2019 /* do an exponential smoothing */
2020 new_itr = (10 * new_itr * q_vector->itr) /
2021 ((9 * new_itr) + q_vector->itr);
2023 /* save the algorithm value here */
2024 q_vector->itr = new_itr;
2026 ixgbe_write_eitr(q_vector);
2031 * ixgbe_check_overtemp_subtask - check for over temperature
2032 * @adapter: pointer to adapter
2034 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2036 struct ixgbe_hw *hw = &adapter->hw;
2037 u32 eicr = adapter->interrupt_event;
2039 if (test_bit(__IXGBE_DOWN, &adapter->state))
2042 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2043 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2046 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2048 switch (hw->device_id) {
2049 case IXGBE_DEV_ID_82599_T3_LOM:
2051 * Since the warning interrupt is for both ports
2052 * we don't have to check if:
2053 * - This interrupt wasn't for our port.
2054 * - We may have missed the interrupt so always have to
2055 * check if we got a LSC
2057 if (!(eicr & IXGBE_EICR_GPI_SDP0) &&
2058 !(eicr & IXGBE_EICR_LSC))
2061 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2063 bool link_up = false;
2065 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2071 /* Check if this is not due to overtemp */
2072 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2077 if (!(eicr & IXGBE_EICR_GPI_SDP0))
2082 "Network adapter has been stopped because it has over heated. "
2083 "Restart the computer. If the problem persists, "
2084 "power off the system and replace the adapter\n");
2086 adapter->interrupt_event = 0;
2089 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2091 struct ixgbe_hw *hw = &adapter->hw;
2093 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2094 (eicr & IXGBE_EICR_GPI_SDP1)) {
2095 e_crit(probe, "Fan has stopped, replace the adapter\n");
2096 /* write to clear the interrupt */
2097 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2101 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2103 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2106 switch (adapter->hw.mac.type) {
2107 case ixgbe_mac_82599EB:
2109 * Need to check link state so complete overtemp check
2112 if (((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)) &&
2113 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2114 adapter->interrupt_event = eicr;
2115 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2116 ixgbe_service_event_schedule(adapter);
2120 case ixgbe_mac_X540:
2121 if (!(eicr & IXGBE_EICR_TS))
2129 "Network adapter has been stopped because it has over heated. "
2130 "Restart the computer. If the problem persists, "
2131 "power off the system and replace the adapter\n");
2134 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2136 struct ixgbe_hw *hw = &adapter->hw;
2138 if (eicr & IXGBE_EICR_GPI_SDP2) {
2139 /* Clear the interrupt */
2140 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
2141 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2142 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2143 ixgbe_service_event_schedule(adapter);
2147 if (eicr & IXGBE_EICR_GPI_SDP1) {
2148 /* Clear the interrupt */
2149 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
2150 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2151 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2152 ixgbe_service_event_schedule(adapter);
2157 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2159 struct ixgbe_hw *hw = &adapter->hw;
2162 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2163 adapter->link_check_timeout = jiffies;
2164 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2165 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2166 IXGBE_WRITE_FLUSH(hw);
2167 ixgbe_service_event_schedule(adapter);
2171 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2175 struct ixgbe_hw *hw = &adapter->hw;
2177 switch (hw->mac.type) {
2178 case ixgbe_mac_82598EB:
2179 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2180 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2182 case ixgbe_mac_82599EB:
2183 case ixgbe_mac_X540:
2184 mask = (qmask & 0xFFFFFFFF);
2186 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2187 mask = (qmask >> 32);
2189 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2194 /* skip the flush */
2197 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2201 struct ixgbe_hw *hw = &adapter->hw;
2203 switch (hw->mac.type) {
2204 case ixgbe_mac_82598EB:
2205 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2206 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2208 case ixgbe_mac_82599EB:
2209 case ixgbe_mac_X540:
2210 mask = (qmask & 0xFFFFFFFF);
2212 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2213 mask = (qmask >> 32);
2215 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2220 /* skip the flush */
2224 * ixgbe_irq_enable - Enable default interrupt generation settings
2225 * @adapter: board private structure
2227 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2230 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2232 /* don't reenable LSC while waiting for link */
2233 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2234 mask &= ~IXGBE_EIMS_LSC;
2236 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2237 switch (adapter->hw.mac.type) {
2238 case ixgbe_mac_82599EB:
2239 mask |= IXGBE_EIMS_GPI_SDP0;
2241 case ixgbe_mac_X540:
2242 mask |= IXGBE_EIMS_TS;
2247 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2248 mask |= IXGBE_EIMS_GPI_SDP1;
2249 switch (adapter->hw.mac.type) {
2250 case ixgbe_mac_82599EB:
2251 mask |= IXGBE_EIMS_GPI_SDP1;
2252 mask |= IXGBE_EIMS_GPI_SDP2;
2253 case ixgbe_mac_X540:
2254 mask |= IXGBE_EIMS_ECC;
2255 mask |= IXGBE_EIMS_MAILBOX;
2260 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2261 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2262 mask |= IXGBE_EIMS_FLOW_DIR;
2264 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2266 ixgbe_irq_enable_queues(adapter, ~0);
2268 IXGBE_WRITE_FLUSH(&adapter->hw);
2271 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2273 struct ixgbe_adapter *adapter = data;
2274 struct ixgbe_hw *hw = &adapter->hw;
2278 * Workaround for Silicon errata. Use clear-by-write instead
2279 * of clear-by-read. Reading with EICS will return the
2280 * interrupt causes without clearing, which later be done
2281 * with the write to EICR.
2283 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2284 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2286 if (eicr & IXGBE_EICR_LSC)
2287 ixgbe_check_lsc(adapter);
2289 if (eicr & IXGBE_EICR_MAILBOX)
2290 ixgbe_msg_task(adapter);
2292 switch (hw->mac.type) {
2293 case ixgbe_mac_82599EB:
2294 case ixgbe_mac_X540:
2295 if (eicr & IXGBE_EICR_ECC)
2296 e_info(link, "Received unrecoverable ECC Err, please "
2298 /* Handle Flow Director Full threshold interrupt */
2299 if (eicr & IXGBE_EICR_FLOW_DIR) {
2300 int reinit_count = 0;
2302 for (i = 0; i < adapter->num_tx_queues; i++) {
2303 struct ixgbe_ring *ring = adapter->tx_ring[i];
2304 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2309 /* no more flow director interrupts until after init */
2310 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2311 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2312 ixgbe_service_event_schedule(adapter);
2315 ixgbe_check_sfp_event(adapter, eicr);
2316 ixgbe_check_overtemp_event(adapter, eicr);
2322 ixgbe_check_fan_failure(adapter, eicr);
2323 #ifdef CONFIG_IXGBE_PTP
2324 ixgbe_ptp_check_pps_event(adapter, eicr);
2327 /* re-enable the original interrupt state, no lsc, no queues */
2328 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2329 ixgbe_irq_enable(adapter, false, false);
2334 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2336 struct ixgbe_q_vector *q_vector = data;
2338 /* EIAM disabled interrupts (on this vector) for us */
2340 if (q_vector->rx.ring || q_vector->tx.ring)
2341 napi_schedule(&q_vector->napi);
2347 * ixgbe_poll - NAPI Rx polling callback
2348 * @napi: structure for representing this polling device
2349 * @budget: how many packets driver is allowed to clean
2351 * This function is used for legacy and MSI, NAPI mode
2353 int ixgbe_poll(struct napi_struct *napi, int budget)
2355 struct ixgbe_q_vector *q_vector =
2356 container_of(napi, struct ixgbe_q_vector, napi);
2357 struct ixgbe_adapter *adapter = q_vector->adapter;
2358 struct ixgbe_ring *ring;
2359 int per_ring_budget;
2360 bool clean_complete = true;
2362 #ifdef CONFIG_IXGBE_DCA
2363 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2364 ixgbe_update_dca(q_vector);
2367 ixgbe_for_each_ring(ring, q_vector->tx)
2368 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2370 /* attempt to distribute budget to each queue fairly, but don't allow
2371 * the budget to go below 1 because we'll exit polling */
2372 if (q_vector->rx.count > 1)
2373 per_ring_budget = max(budget/q_vector->rx.count, 1);
2375 per_ring_budget = budget;
2377 ixgbe_for_each_ring(ring, q_vector->rx)
2378 clean_complete &= ixgbe_clean_rx_irq(q_vector, ring,
2381 /* If all work not completed, return budget and keep polling */
2382 if (!clean_complete)
2385 /* all work done, exit the polling mode */
2386 napi_complete(napi);
2387 if (adapter->rx_itr_setting & 1)
2388 ixgbe_set_itr(q_vector);
2389 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2390 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2396 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2397 * @adapter: board private structure
2399 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2400 * interrupts from the kernel.
2402 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2404 struct net_device *netdev = adapter->netdev;
2408 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2409 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2410 struct msix_entry *entry = &adapter->msix_entries[vector];
2412 if (q_vector->tx.ring && q_vector->rx.ring) {
2413 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2414 "%s-%s-%d", netdev->name, "TxRx", ri++);
2416 } else if (q_vector->rx.ring) {
2417 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2418 "%s-%s-%d", netdev->name, "rx", ri++);
2419 } else if (q_vector->tx.ring) {
2420 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2421 "%s-%s-%d", netdev->name, "tx", ti++);
2423 /* skip this unused q_vector */
2426 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2427 q_vector->name, q_vector);
2429 e_err(probe, "request_irq failed for MSIX interrupt "
2430 "Error: %d\n", err);
2431 goto free_queue_irqs;
2433 /* If Flow Director is enabled, set interrupt affinity */
2434 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2435 /* assign the mask for this irq */
2436 irq_set_affinity_hint(entry->vector,
2437 &q_vector->affinity_mask);
2441 err = request_irq(adapter->msix_entries[vector].vector,
2442 ixgbe_msix_other, 0, netdev->name, adapter);
2444 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2445 goto free_queue_irqs;
2453 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2455 free_irq(adapter->msix_entries[vector].vector,
2456 adapter->q_vector[vector]);
2458 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2459 pci_disable_msix(adapter->pdev);
2460 kfree(adapter->msix_entries);
2461 adapter->msix_entries = NULL;
2466 * ixgbe_intr - legacy mode Interrupt Handler
2467 * @irq: interrupt number
2468 * @data: pointer to a network interface device structure
2470 static irqreturn_t ixgbe_intr(int irq, void *data)
2472 struct ixgbe_adapter *adapter = data;
2473 struct ixgbe_hw *hw = &adapter->hw;
2474 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2478 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2479 * before the read of EICR.
2481 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2483 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2484 * therefore no explicit interrupt disable is necessary */
2485 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2488 * shared interrupt alert!
2489 * make sure interrupts are enabled because the read will
2490 * have disabled interrupts due to EIAM
2491 * finish the workaround of silicon errata on 82598. Unmask
2492 * the interrupt that we masked before the EICR read.
2494 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2495 ixgbe_irq_enable(adapter, true, true);
2496 return IRQ_NONE; /* Not our interrupt */
2499 if (eicr & IXGBE_EICR_LSC)
2500 ixgbe_check_lsc(adapter);
2502 switch (hw->mac.type) {
2503 case ixgbe_mac_82599EB:
2504 ixgbe_check_sfp_event(adapter, eicr);
2506 case ixgbe_mac_X540:
2507 if (eicr & IXGBE_EICR_ECC)
2508 e_info(link, "Received unrecoverable ECC err, please "
2510 ixgbe_check_overtemp_event(adapter, eicr);
2516 ixgbe_check_fan_failure(adapter, eicr);
2517 #ifdef CONFIG_IXGBE_PTP
2518 ixgbe_ptp_check_pps_event(adapter, eicr);
2521 /* would disable interrupts here but EIAM disabled it */
2522 napi_schedule(&q_vector->napi);
2525 * re-enable link(maybe) and non-queue interrupts, no flush.
2526 * ixgbe_poll will re-enable the queue interrupts
2528 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2529 ixgbe_irq_enable(adapter, false, false);
2535 * ixgbe_request_irq - initialize interrupts
2536 * @adapter: board private structure
2538 * Attempts to configure interrupts using the best available
2539 * capabilities of the hardware and kernel.
2541 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2543 struct net_device *netdev = adapter->netdev;
2546 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2547 err = ixgbe_request_msix_irqs(adapter);
2548 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2549 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2550 netdev->name, adapter);
2552 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2553 netdev->name, adapter);
2556 e_err(probe, "request_irq failed, Error %d\n", err);
2561 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2565 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2566 free_irq(adapter->pdev->irq, adapter);
2570 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2571 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2572 struct msix_entry *entry = &adapter->msix_entries[vector];
2574 /* free only the irqs that were actually requested */
2575 if (!q_vector->rx.ring && !q_vector->tx.ring)
2578 /* clear the affinity_mask in the IRQ descriptor */
2579 irq_set_affinity_hint(entry->vector, NULL);
2581 free_irq(entry->vector, q_vector);
2584 free_irq(adapter->msix_entries[vector++].vector, adapter);
2588 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2589 * @adapter: board private structure
2591 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2593 switch (adapter->hw.mac.type) {
2594 case ixgbe_mac_82598EB:
2595 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2597 case ixgbe_mac_82599EB:
2598 case ixgbe_mac_X540:
2599 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2600 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2601 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2606 IXGBE_WRITE_FLUSH(&adapter->hw);
2607 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2610 for (vector = 0; vector < adapter->num_q_vectors; vector++)
2611 synchronize_irq(adapter->msix_entries[vector].vector);
2613 synchronize_irq(adapter->msix_entries[vector++].vector);
2615 synchronize_irq(adapter->pdev->irq);
2620 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2623 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2625 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2628 if (adapter->rx_itr_setting == 1)
2629 q_vector->itr = IXGBE_20K_ITR;
2631 q_vector->itr = adapter->rx_itr_setting;
2633 ixgbe_write_eitr(q_vector);
2635 ixgbe_set_ivar(adapter, 0, 0, 0);
2636 ixgbe_set_ivar(adapter, 1, 0, 0);
2638 e_info(hw, "Legacy interrupt IVAR setup done\n");
2642 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2643 * @adapter: board private structure
2644 * @ring: structure containing ring specific data
2646 * Configure the Tx descriptor ring after a reset.
2648 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2649 struct ixgbe_ring *ring)
2651 struct ixgbe_hw *hw = &adapter->hw;
2652 u64 tdba = ring->dma;
2654 u32 txdctl = IXGBE_TXDCTL_ENABLE;
2655 u8 reg_idx = ring->reg_idx;
2657 /* disable queue to avoid issues while updating state */
2658 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
2659 IXGBE_WRITE_FLUSH(hw);
2661 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2662 (tdba & DMA_BIT_MASK(32)));
2663 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2664 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2665 ring->count * sizeof(union ixgbe_adv_tx_desc));
2666 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2667 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2668 ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2671 * set WTHRESH to encourage burst writeback, it should not be set
2672 * higher than 1 when ITR is 0 as it could cause false TX hangs
2674 * In order to avoid issues WTHRESH + PTHRESH should always be equal
2675 * to or less than the number of on chip descriptors, which is
2678 if (!ring->q_vector || (ring->q_vector->itr < 8))
2679 txdctl |= (1 << 16); /* WTHRESH = 1 */
2681 txdctl |= (8 << 16); /* WTHRESH = 8 */
2684 * Setting PTHRESH to 32 both improves performance
2685 * and avoids a TX hang with DFP enabled
2687 txdctl |= (1 << 8) | /* HTHRESH = 1 */
2688 32; /* PTHRESH = 32 */
2690 /* reinitialize flowdirector state */
2691 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2692 adapter->atr_sample_rate) {
2693 ring->atr_sample_rate = adapter->atr_sample_rate;
2694 ring->atr_count = 0;
2695 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2697 ring->atr_sample_rate = 0;
2700 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2703 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2705 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2706 if (hw->mac.type == ixgbe_mac_82598EB &&
2707 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2710 /* poll to verify queue is enabled */
2712 usleep_range(1000, 2000);
2713 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2714 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2716 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2719 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2721 struct ixgbe_hw *hw = &adapter->hw;
2724 u8 tcs = netdev_get_num_tc(adapter->netdev);
2726 if (hw->mac.type == ixgbe_mac_82598EB)
2729 /* disable the arbiter while setting MTQC */
2730 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2731 rttdcs |= IXGBE_RTTDCS_ARBDIS;
2732 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2734 /* set transmit pool layout */
2735 switch (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
2736 case (IXGBE_FLAG_SRIOV_ENABLED):
2737 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2738 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2742 reg = IXGBE_MTQC_64Q_1PB;
2744 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
2746 reg = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
2748 IXGBE_WRITE_REG(hw, IXGBE_MTQC, reg);
2750 /* Enable Security TX Buffer IFG for multiple pb */
2752 reg = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
2753 reg |= IXGBE_SECTX_DCB;
2754 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, reg);
2759 /* re-enable the arbiter */
2760 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2761 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2765 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2766 * @adapter: board private structure
2768 * Configure the Tx unit of the MAC after a reset.
2770 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2772 struct ixgbe_hw *hw = &adapter->hw;
2776 ixgbe_setup_mtqc(adapter);
2778 if (hw->mac.type != ixgbe_mac_82598EB) {
2779 /* DMATXCTL.EN must be before Tx queues are enabled */
2780 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2781 dmatxctl |= IXGBE_DMATXCTL_TE;
2782 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2785 /* Setup the HW Tx Head and Tail descriptor pointers */
2786 for (i = 0; i < adapter->num_tx_queues; i++)
2787 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2790 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
2791 struct ixgbe_ring *ring)
2793 struct ixgbe_hw *hw = &adapter->hw;
2794 u8 reg_idx = ring->reg_idx;
2795 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2797 srrctl |= IXGBE_SRRCTL_DROP_EN;
2799 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2802 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
2803 struct ixgbe_ring *ring)
2805 struct ixgbe_hw *hw = &adapter->hw;
2806 u8 reg_idx = ring->reg_idx;
2807 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
2809 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
2811 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
2814 #ifdef CONFIG_IXGBE_DCB
2815 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2817 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
2821 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
2823 if (adapter->ixgbe_ieee_pfc)
2824 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
2827 * We should set the drop enable bit if:
2830 * Number of Rx queues > 1 and flow control is disabled
2832 * This allows us to avoid head of line blocking for security
2833 * and performance reasons.
2835 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
2836 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
2837 for (i = 0; i < adapter->num_rx_queues; i++)
2838 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
2840 for (i = 0; i < adapter->num_rx_queues; i++)
2841 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
2845 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2847 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2848 struct ixgbe_ring *rx_ring)
2851 u8 reg_idx = rx_ring->reg_idx;
2853 switch (adapter->hw.mac.type) {
2854 case ixgbe_mac_82598EB: {
2855 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2856 const int mask = feature[RING_F_RSS].mask;
2857 reg_idx = reg_idx & mask;
2860 case ixgbe_mac_82599EB:
2861 case ixgbe_mac_X540:
2866 srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2868 srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2869 srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2870 if (adapter->num_vfs)
2871 srrctl |= IXGBE_SRRCTL_DROP_EN;
2873 srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2874 IXGBE_SRRCTL_BSIZEHDR_MASK;
2876 #if PAGE_SIZE > IXGBE_MAX_RXBUFFER
2877 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2879 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2881 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2883 IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2886 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2888 struct ixgbe_hw *hw = &adapter->hw;
2889 static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2890 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2891 0x6A3E67EA, 0x14364D17, 0x3BED200D};
2892 u32 mrqc = 0, reta = 0;
2895 u8 tcs = netdev_get_num_tc(adapter->netdev);
2896 int maxq = adapter->ring_feature[RING_F_RSS].indices;
2899 maxq = min(maxq, adapter->num_tx_queues / tcs);
2901 /* Fill out hash function seeds */
2902 for (i = 0; i < 10; i++)
2903 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2905 /* Fill out redirection table */
2906 for (i = 0, j = 0; i < 128; i++, j++) {
2909 /* reta = 4-byte sliding window of
2910 * 0x00..(indices-1)(indices-1)00..etc. */
2911 reta = (reta << 8) | (j * 0x11);
2913 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2916 /* Disable indicating checksum in descriptor, enables RSS hash */
2917 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2918 rxcsum |= IXGBE_RXCSUM_PCSD;
2919 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2921 if (adapter->hw.mac.type == ixgbe_mac_82598EB &&
2922 (adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
2923 mrqc = IXGBE_MRQC_RSSEN;
2925 int mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2926 | IXGBE_FLAG_SRIOV_ENABLED);
2929 case (IXGBE_FLAG_RSS_ENABLED):
2931 mrqc = IXGBE_MRQC_RSSEN;
2933 mrqc = IXGBE_MRQC_RTRSS4TCEN;
2935 mrqc = IXGBE_MRQC_RTRSS8TCEN;
2937 case (IXGBE_FLAG_SRIOV_ENABLED):
2938 mrqc = IXGBE_MRQC_VMDQEN;
2945 /* Perform hash on these packet types */
2946 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2947 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2948 | IXGBE_MRQC_RSS_FIELD_IPV6
2949 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2951 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
2952 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
2953 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
2954 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2956 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2960 * ixgbe_configure_rscctl - enable RSC for the indicated ring
2961 * @adapter: address of board private structure
2962 * @index: index of ring to set
2964 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2965 struct ixgbe_ring *ring)
2967 struct ixgbe_hw *hw = &adapter->hw;
2969 u8 reg_idx = ring->reg_idx;
2971 if (!ring_is_rsc_enabled(ring))
2974 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2975 rscctrl |= IXGBE_RSCCTL_RSCEN;
2977 * we must limit the number of descriptors so that the
2978 * total size of max desc * buf_len is not greater
2981 #if (PAGE_SIZE <= 8192)
2982 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2983 #elif (PAGE_SIZE <= 16384)
2984 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2986 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2988 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2991 #define IXGBE_MAX_RX_DESC_POLL 10
2992 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2993 struct ixgbe_ring *ring)
2995 struct ixgbe_hw *hw = &adapter->hw;
2996 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2998 u8 reg_idx = ring->reg_idx;
3000 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3001 if (hw->mac.type == ixgbe_mac_82598EB &&
3002 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3006 usleep_range(1000, 2000);
3007 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3008 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3011 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3012 "the polling period\n", reg_idx);
3016 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3017 struct ixgbe_ring *ring)
3019 struct ixgbe_hw *hw = &adapter->hw;
3020 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3022 u8 reg_idx = ring->reg_idx;
3024 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3025 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3027 /* write value back with RXDCTL.ENABLE bit cleared */
3028 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3030 if (hw->mac.type == ixgbe_mac_82598EB &&
3031 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3034 /* the hardware may take up to 100us to really disable the rx queue */
3037 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3038 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3041 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3042 "the polling period\n", reg_idx);
3046 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3047 struct ixgbe_ring *ring)
3049 struct ixgbe_hw *hw = &adapter->hw;
3050 u64 rdba = ring->dma;
3052 u8 reg_idx = ring->reg_idx;
3054 /* disable queue to avoid issues while updating state */
3055 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3056 ixgbe_disable_rx_queue(adapter, ring);
3058 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3059 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3060 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3061 ring->count * sizeof(union ixgbe_adv_rx_desc));
3062 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3063 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3064 ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3066 ixgbe_configure_srrctl(adapter, ring);
3067 ixgbe_configure_rscctl(adapter, ring);
3069 /* If operating in IOV mode set RLPML for X540 */
3070 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
3071 hw->mac.type == ixgbe_mac_X540) {
3072 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
3073 rxdctl |= ((ring->netdev->mtu + ETH_HLEN +
3074 ETH_FCS_LEN + VLAN_HLEN) | IXGBE_RXDCTL_RLPML_EN);
3077 if (hw->mac.type == ixgbe_mac_82598EB) {
3079 * enable cache line friendly hardware writes:
3080 * PTHRESH=32 descriptors (half the internal cache),
3081 * this also removes ugly rx_no_buffer_count increment
3082 * HTHRESH=4 descriptors (to minimize latency on fetch)
3083 * WTHRESH=8 burst writeback up to two cache lines
3085 rxdctl &= ~0x3FFFFF;
3089 /* enable receive descriptor ring */
3090 rxdctl |= IXGBE_RXDCTL_ENABLE;
3091 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3093 ixgbe_rx_desc_queue_enable(adapter, ring);
3094 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3097 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3099 struct ixgbe_hw *hw = &adapter->hw;
3102 /* PSRTYPE must be initialized in non 82598 adapters */
3103 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3104 IXGBE_PSRTYPE_UDPHDR |
3105 IXGBE_PSRTYPE_IPV4HDR |
3106 IXGBE_PSRTYPE_L2HDR |
3107 IXGBE_PSRTYPE_IPV6HDR;
3109 if (hw->mac.type == ixgbe_mac_82598EB)
3112 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3113 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3115 for (p = 0; p < adapter->num_rx_pools; p++)
3116 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3120 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3122 struct ixgbe_hw *hw = &adapter->hw;
3125 u32 reg_offset, vf_shift;
3129 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3132 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3133 vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3134 vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3135 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3137 vf_shift = adapter->num_vfs % 32;
3138 reg_offset = (adapter->num_vfs >= 32) ? 1 : 0;
3140 /* Enable only the PF's pool for Tx/Rx */
3141 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3142 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3143 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3144 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3145 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3147 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3148 hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3151 * Set up VF register offsets for selected VT Mode,
3152 * i.e. 32 or 64 VFs for SR-IOV
3154 gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3155 gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3156 gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3157 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3159 /* enable Tx loopback for VF/PF communication */
3160 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3161 /* Enable MAC Anti-Spoofing */
3162 hw->mac.ops.set_mac_anti_spoofing(hw,
3163 (adapter->num_vfs != 0),
3165 /* For VFs that have spoof checking turned off */
3166 for (i = 0; i < adapter->num_vfs; i++) {
3167 if (!adapter->vfinfo[i].spoofchk_enabled)
3168 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3172 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3174 struct ixgbe_hw *hw = &adapter->hw;
3175 struct net_device *netdev = adapter->netdev;
3176 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3177 struct ixgbe_ring *rx_ring;
3182 /* adjust max frame to be able to do baby jumbo for FCoE */
3183 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3184 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3185 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3187 #endif /* IXGBE_FCOE */
3188 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3189 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3190 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3191 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3193 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3196 /* MHADD will allow an extra 4 bytes past for vlan tagged frames */
3197 max_frame += VLAN_HLEN;
3199 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3200 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3201 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3202 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3205 * Setup the HW Rx Head and Tail Descriptor Pointers and
3206 * the Base and Length of the Rx Descriptor Ring
3208 for (i = 0; i < adapter->num_rx_queues; i++) {
3209 rx_ring = adapter->rx_ring[i];
3210 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3211 set_ring_rsc_enabled(rx_ring);
3213 clear_ring_rsc_enabled(rx_ring);
3217 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3219 struct ixgbe_hw *hw = &adapter->hw;
3220 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3222 switch (hw->mac.type) {
3223 case ixgbe_mac_82598EB:
3225 * For VMDq support of different descriptor types or
3226 * buffer sizes through the use of multiple SRRCTL
3227 * registers, RDRXCTL.MVMEN must be set to 1
3229 * also, the manual doesn't mention it clearly but DCA hints
3230 * will only use queue 0's tags unless this bit is set. Side
3231 * effects of setting this bit are only that SRRCTL must be
3232 * fully programmed [0..15]
3234 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3236 case ixgbe_mac_82599EB:
3237 case ixgbe_mac_X540:
3238 /* Disable RSC for ACK packets */
3239 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3240 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3241 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3242 /* hardware requires some bits to be set by default */
3243 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3244 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3247 /* We should do nothing since we don't know this hardware */
3251 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3255 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3256 * @adapter: board private structure
3258 * Configure the Rx unit of the MAC after a reset.
3260 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3262 struct ixgbe_hw *hw = &adapter->hw;
3266 /* disable receives while setting up the descriptors */
3267 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3268 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3270 ixgbe_setup_psrtype(adapter);
3271 ixgbe_setup_rdrxctl(adapter);
3273 /* Program registers for the distribution of queues */
3274 ixgbe_setup_mrqc(adapter);
3276 /* set_rx_buffer_len must be called before ring initialization */
3277 ixgbe_set_rx_buffer_len(adapter);
3280 * Setup the HW Rx Head and Tail Descriptor Pointers and
3281 * the Base and Length of the Rx Descriptor Ring
3283 for (i = 0; i < adapter->num_rx_queues; i++)
3284 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3286 /* disable drop enable for 82598 parts */
3287 if (hw->mac.type == ixgbe_mac_82598EB)
3288 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3290 /* enable all receives */
3291 rxctrl |= IXGBE_RXCTRL_RXEN;
3292 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3295 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3297 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3298 struct ixgbe_hw *hw = &adapter->hw;
3299 int pool_ndx = adapter->num_vfs;
3301 /* add VID to filter table */
3302 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3303 set_bit(vid, adapter->active_vlans);
3308 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3310 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3311 struct ixgbe_hw *hw = &adapter->hw;
3312 int pool_ndx = adapter->num_vfs;
3314 /* remove VID from filter table */
3315 hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3316 clear_bit(vid, adapter->active_vlans);
3322 * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3323 * @adapter: driver data
3325 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3327 struct ixgbe_hw *hw = &adapter->hw;
3330 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3331 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3332 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3336 * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3337 * @adapter: driver data
3339 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3341 struct ixgbe_hw *hw = &adapter->hw;
3344 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3345 vlnctrl |= IXGBE_VLNCTRL_VFE;
3346 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3347 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3351 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3352 * @adapter: driver data
3354 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3356 struct ixgbe_hw *hw = &adapter->hw;
3360 switch (hw->mac.type) {
3361 case ixgbe_mac_82598EB:
3362 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3363 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3364 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3366 case ixgbe_mac_82599EB:
3367 case ixgbe_mac_X540:
3368 for (i = 0; i < adapter->num_rx_queues; i++) {
3369 j = adapter->rx_ring[i]->reg_idx;
3370 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3371 vlnctrl &= ~IXGBE_RXDCTL_VME;
3372 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3381 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3382 * @adapter: driver data
3384 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3386 struct ixgbe_hw *hw = &adapter->hw;
3390 switch (hw->mac.type) {
3391 case ixgbe_mac_82598EB:
3392 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3393 vlnctrl |= IXGBE_VLNCTRL_VME;
3394 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3396 case ixgbe_mac_82599EB:
3397 case ixgbe_mac_X540:
3398 for (i = 0; i < adapter->num_rx_queues; i++) {
3399 j = adapter->rx_ring[i]->reg_idx;
3400 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3401 vlnctrl |= IXGBE_RXDCTL_VME;
3402 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3410 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3414 ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3416 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3417 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3421 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3422 * @netdev: network interface device structure
3424 * Writes unicast address list to the RAR table.
3425 * Returns: -ENOMEM on failure/insufficient address space
3426 * 0 on no addresses written
3427 * X on writing X addresses to the RAR table
3429 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3431 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3432 struct ixgbe_hw *hw = &adapter->hw;
3433 unsigned int vfn = adapter->num_vfs;
3434 unsigned int rar_entries = IXGBE_MAX_PF_MACVLANS;
3437 /* return ENOMEM indicating insufficient memory for addresses */
3438 if (netdev_uc_count(netdev) > rar_entries)
3441 if (!netdev_uc_empty(netdev) && rar_entries) {
3442 struct netdev_hw_addr *ha;
3443 /* return error if we do not support writing to RAR table */
3444 if (!hw->mac.ops.set_rar)
3447 netdev_for_each_uc_addr(ha, netdev) {
3450 hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3455 /* write the addresses in reverse order to avoid write combining */
3456 for (; rar_entries > 0 ; rar_entries--)
3457 hw->mac.ops.clear_rar(hw, rar_entries);
3463 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3464 * @netdev: network interface device structure
3466 * The set_rx_method entry point is called whenever the unicast/multicast
3467 * address list or the network interface flags are updated. This routine is
3468 * responsible for configuring the hardware for proper unicast, multicast and
3471 void ixgbe_set_rx_mode(struct net_device *netdev)
3473 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3474 struct ixgbe_hw *hw = &adapter->hw;
3475 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3478 /* Check for Promiscuous and All Multicast modes */
3480 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3482 /* set all bits that we expect to always be set */
3483 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
3484 fctrl |= IXGBE_FCTRL_BAM;
3485 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3486 fctrl |= IXGBE_FCTRL_PMCF;
3488 /* clear the bits we are changing the status of */
3489 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3491 if (netdev->flags & IFF_PROMISC) {
3492 hw->addr_ctrl.user_set_promisc = true;
3493 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3494 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3495 /* don't hardware filter vlans in promisc mode */
3496 ixgbe_vlan_filter_disable(adapter);
3498 if (netdev->flags & IFF_ALLMULTI) {
3499 fctrl |= IXGBE_FCTRL_MPE;
3500 vmolr |= IXGBE_VMOLR_MPE;
3503 * Write addresses to the MTA, if the attempt fails
3504 * then we should just turn on promiscuous mode so
3505 * that we can at least receive multicast traffic
3507 hw->mac.ops.update_mc_addr_list(hw, netdev);
3508 vmolr |= IXGBE_VMOLR_ROMPE;
3510 ixgbe_vlan_filter_enable(adapter);
3511 hw->addr_ctrl.user_set_promisc = false;
3515 * Write addresses to available RAR registers, if there is not
3516 * sufficient space to store all the addresses then enable
3517 * unicast promiscuous mode
3519 count = ixgbe_write_uc_addr_list(netdev);
3521 fctrl |= IXGBE_FCTRL_UPE;
3522 vmolr |= IXGBE_VMOLR_ROPE;
3525 if (adapter->num_vfs) {
3526 ixgbe_restore_vf_multicasts(adapter);
3527 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3528 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3530 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3533 /* This is useful for sniffing bad packets. */
3534 if (adapter->netdev->features & NETIF_F_RXALL) {
3535 /* UPE and MPE will be handled by normal PROMISC logic
3536 * in e1000e_set_rx_mode */
3537 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
3538 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
3539 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
3541 fctrl &= ~(IXGBE_FCTRL_DPF);
3542 /* NOTE: VLAN filtering is disabled by setting PROMISC */
3545 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3547 if (netdev->features & NETIF_F_HW_VLAN_RX)
3548 ixgbe_vlan_strip_enable(adapter);
3550 ixgbe_vlan_strip_disable(adapter);
3553 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3557 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3558 napi_enable(&adapter->q_vector[q_idx]->napi);
3561 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3565 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
3566 napi_disable(&adapter->q_vector[q_idx]->napi);
3569 #ifdef CONFIG_IXGBE_DCB
3571 * ixgbe_configure_dcb - Configure DCB hardware
3572 * @adapter: ixgbe adapter struct
3574 * This is called by the driver on open to configure the DCB hardware.
3575 * This is also called by the gennetlink interface when reconfiguring
3578 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3580 struct ixgbe_hw *hw = &adapter->hw;
3581 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3583 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3584 if (hw->mac.type == ixgbe_mac_82598EB)
3585 netif_set_gso_max_size(adapter->netdev, 65536);
3589 if (hw->mac.type == ixgbe_mac_82598EB)
3590 netif_set_gso_max_size(adapter->netdev, 32768);
3592 hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3595 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3596 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3599 /* reconfigure the hardware */
3600 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
3601 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3603 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3605 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3606 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
3607 ixgbe_dcb_hw_ets(&adapter->hw,
3608 adapter->ixgbe_ieee_ets,
3610 ixgbe_dcb_hw_pfc_config(&adapter->hw,
3611 adapter->ixgbe_ieee_pfc->pfc_en,
3612 adapter->ixgbe_ieee_ets->prio_tc);
3615 /* Enable RSS Hash per TC */
3616 if (hw->mac.type != ixgbe_mac_82598EB) {
3620 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
3622 u8 cnt = adapter->netdev->tc_to_txq[i].count;
3627 reg |= msb << IXGBE_RQTC_SHIFT_TC(i);
3629 IXGBE_WRITE_REG(hw, IXGBE_RQTC, reg);
3634 /* Additional bittime to account for IXGBE framing */
3635 #define IXGBE_ETH_FRAMING 20
3638 * ixgbe_hpbthresh - calculate high water mark for flow control
3640 * @adapter: board private structure to calculate for
3641 * @pb: packet buffer to calculate
3643 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
3645 struct ixgbe_hw *hw = &adapter->hw;
3646 struct net_device *dev = adapter->netdev;
3647 int link, tc, kb, marker;
3650 /* Calculate max LAN frame size */
3651 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
3654 /* FCoE traffic class uses FCOE jumbo frames */
3655 if (dev->features & NETIF_F_FCOE_MTU) {
3658 #ifdef CONFIG_IXGBE_DCB
3659 fcoe_pb = netdev_get_prio_tc_map(dev, adapter->fcoe.up);
3662 if (fcoe_pb == pb && tc < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3663 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3667 /* Calculate delay value for device */
3668 switch (hw->mac.type) {
3669 case ixgbe_mac_X540:
3670 dv_id = IXGBE_DV_X540(link, tc);
3673 dv_id = IXGBE_DV(link, tc);
3677 /* Loopback switch introduces additional latency */
3678 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3679 dv_id += IXGBE_B2BT(tc);
3681 /* Delay value is calculated in bit times convert to KB */
3682 kb = IXGBE_BT2KB(dv_id);
3683 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
3685 marker = rx_pba - kb;
3687 /* It is possible that the packet buffer is not large enough
3688 * to provide required headroom. In this case throw an error
3689 * to user and a do the best we can.
3692 e_warn(drv, "Packet Buffer(%i) can not provide enough"
3693 "headroom to support flow control."
3694 "Decrease MTU or number of traffic classes\n", pb);
3702 * ixgbe_lpbthresh - calculate low water mark for for flow control
3704 * @adapter: board private structure to calculate for
3705 * @pb: packet buffer to calculate
3707 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter)
3709 struct ixgbe_hw *hw = &adapter->hw;
3710 struct net_device *dev = adapter->netdev;
3714 /* Calculate max LAN frame size */
3715 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
3717 /* Calculate delay value for device */
3718 switch (hw->mac.type) {
3719 case ixgbe_mac_X540:
3720 dv_id = IXGBE_LOW_DV_X540(tc);
3723 dv_id = IXGBE_LOW_DV(tc);
3727 /* Delay value is calculated in bit times convert to KB */
3728 return IXGBE_BT2KB(dv_id);
3732 * ixgbe_pbthresh_setup - calculate and setup high low water marks
3734 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
3736 struct ixgbe_hw *hw = &adapter->hw;
3737 int num_tc = netdev_get_num_tc(adapter->netdev);
3743 hw->fc.low_water = ixgbe_lpbthresh(adapter);
3745 for (i = 0; i < num_tc; i++) {
3746 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
3748 /* Low water marks must not be larger than high water marks */
3749 if (hw->fc.low_water > hw->fc.high_water[i])
3750 hw->fc.low_water = 0;
3754 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
3756 struct ixgbe_hw *hw = &adapter->hw;
3758 u8 tc = netdev_get_num_tc(adapter->netdev);
3760 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3761 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3762 hdrm = 32 << adapter->fdir_pballoc;
3766 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
3767 ixgbe_pbthresh_setup(adapter);
3770 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
3772 struct ixgbe_hw *hw = &adapter->hw;
3773 struct hlist_node *node, *node2;
3774 struct ixgbe_fdir_filter *filter;
3776 spin_lock(&adapter->fdir_perfect_lock);
3778 if (!hlist_empty(&adapter->fdir_filter_list))
3779 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
3781 hlist_for_each_entry_safe(filter, node, node2,
3782 &adapter->fdir_filter_list, fdir_node) {
3783 ixgbe_fdir_write_perfect_filter_82599(hw,
3786 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
3787 IXGBE_FDIR_DROP_QUEUE :
3788 adapter->rx_ring[filter->action]->reg_idx);
3791 spin_unlock(&adapter->fdir_perfect_lock);
3794 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3796 struct ixgbe_hw *hw = &adapter->hw;
3798 ixgbe_configure_pb(adapter);
3799 #ifdef CONFIG_IXGBE_DCB
3800 ixgbe_configure_dcb(adapter);
3803 ixgbe_set_rx_mode(adapter->netdev);
3804 ixgbe_restore_vlan(adapter);
3807 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3808 ixgbe_configure_fcoe(adapter);
3810 #endif /* IXGBE_FCOE */
3812 switch (hw->mac.type) {
3813 case ixgbe_mac_82599EB:
3814 case ixgbe_mac_X540:
3815 hw->mac.ops.disable_rx_buff(hw);
3821 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3822 ixgbe_init_fdir_signature_82599(&adapter->hw,
3823 adapter->fdir_pballoc);
3824 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3825 ixgbe_init_fdir_perfect_82599(&adapter->hw,
3826 adapter->fdir_pballoc);
3827 ixgbe_fdir_filter_restore(adapter);
3830 switch (hw->mac.type) {
3831 case ixgbe_mac_82599EB:
3832 case ixgbe_mac_X540:
3833 hw->mac.ops.enable_rx_buff(hw);
3839 ixgbe_configure_virtualization(adapter);
3841 ixgbe_configure_tx(adapter);
3842 ixgbe_configure_rx(adapter);
3845 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3847 switch (hw->phy.type) {
3848 case ixgbe_phy_sfp_avago:
3849 case ixgbe_phy_sfp_ftl:
3850 case ixgbe_phy_sfp_intel:
3851 case ixgbe_phy_sfp_unknown:
3852 case ixgbe_phy_sfp_passive_tyco:
3853 case ixgbe_phy_sfp_passive_unknown:
3854 case ixgbe_phy_sfp_active_unknown:
3855 case ixgbe_phy_sfp_ftl_active:
3858 if (hw->mac.type == ixgbe_mac_82598EB)
3866 * ixgbe_sfp_link_config - set up SFP+ link
3867 * @adapter: pointer to private adapter struct
3869 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3872 * We are assuming the worst case scenario here, and that
3873 * is that an SFP was inserted/removed after the reset
3874 * but before SFP detection was enabled. As such the best
3875 * solution is to just start searching as soon as we start
3877 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3878 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
3880 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
3884 * ixgbe_non_sfp_link_config - set up non-SFP+ link
3885 * @hw: pointer to private hardware struct
3887 * Returns 0 on success, negative on failure
3889 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3892 bool negotiation, link_up = false;
3893 u32 ret = IXGBE_ERR_LINK_SETUP;
3895 if (hw->mac.ops.check_link)
3896 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3901 autoneg = hw->phy.autoneg_advertised;
3902 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
3903 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3908 if (hw->mac.ops.setup_link)
3909 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3914 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3916 struct ixgbe_hw *hw = &adapter->hw;
3919 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3920 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3922 gpie |= IXGBE_GPIE_EIAME;
3924 * use EIAM to auto-mask when MSI-X interrupt is asserted
3925 * this saves a register write for every interrupt
3927 switch (hw->mac.type) {
3928 case ixgbe_mac_82598EB:
3929 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3931 case ixgbe_mac_82599EB:
3932 case ixgbe_mac_X540:
3934 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3935 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3939 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3940 * specifically only auto mask tx and rx interrupts */
3941 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3944 /* XXX: to interrupt immediately for EICS writes, enable this */
3945 /* gpie |= IXGBE_GPIE_EIMEN; */
3947 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3948 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3949 gpie |= IXGBE_GPIE_VTMODE_64;
3952 /* Enable Thermal over heat sensor interrupt */
3953 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
3954 switch (adapter->hw.mac.type) {
3955 case ixgbe_mac_82599EB:
3956 gpie |= IXGBE_SDP0_GPIEN;
3958 case ixgbe_mac_X540:
3959 gpie |= IXGBE_EIMS_TS;
3966 /* Enable fan failure interrupt */
3967 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3968 gpie |= IXGBE_SDP1_GPIEN;
3970 if (hw->mac.type == ixgbe_mac_82599EB) {
3971 gpie |= IXGBE_SDP1_GPIEN;
3972 gpie |= IXGBE_SDP2_GPIEN;
3975 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3978 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
3980 struct ixgbe_hw *hw = &adapter->hw;
3984 ixgbe_get_hw_control(adapter);
3985 ixgbe_setup_gpie(adapter);
3987 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3988 ixgbe_configure_msix(adapter);
3990 ixgbe_configure_msi_and_legacy(adapter);
3992 /* enable the optics for both mult-speed fiber and 82599 SFP+ fiber */
3993 if (hw->mac.ops.enable_tx_laser &&
3994 ((hw->phy.multispeed_fiber) ||
3995 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
3996 (hw->mac.type == ixgbe_mac_82599EB))))
3997 hw->mac.ops.enable_tx_laser(hw);
3999 clear_bit(__IXGBE_DOWN, &adapter->state);
4000 ixgbe_napi_enable_all(adapter);
4002 if (ixgbe_is_sfp(hw)) {
4003 ixgbe_sfp_link_config(adapter);
4005 err = ixgbe_non_sfp_link_config(hw);
4007 e_err(probe, "link_config FAILED %d\n", err);
4010 /* clear any pending interrupts, may auto mask */
4011 IXGBE_READ_REG(hw, IXGBE_EICR);
4012 ixgbe_irq_enable(adapter, true, true);
4015 * If this adapter has a fan, check to see if we had a failure
4016 * before we enabled the interrupt.
4018 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4019 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4020 if (esdp & IXGBE_ESDP_SDP1)
4021 e_crit(drv, "Fan has stopped, replace the adapter\n");
4024 /* enable transmits */
4025 netif_tx_start_all_queues(adapter->netdev);
4027 /* bring the link up in the watchdog, this could race with our first
4028 * link up interrupt but shouldn't be a problem */
4029 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4030 adapter->link_check_timeout = jiffies;
4031 mod_timer(&adapter->service_timer, jiffies);
4033 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4034 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4035 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4036 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4039 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4041 WARN_ON(in_interrupt());
4042 /* put off any impending NetWatchDogTimeout */
4043 adapter->netdev->trans_start = jiffies;
4045 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4046 usleep_range(1000, 2000);
4047 ixgbe_down(adapter);
4049 * If SR-IOV enabled then wait a bit before bringing the adapter
4050 * back up to give the VFs time to respond to the reset. The
4051 * two second wait is based upon the watchdog timer cycle in
4054 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4057 clear_bit(__IXGBE_RESETTING, &adapter->state);
4060 void ixgbe_up(struct ixgbe_adapter *adapter)
4062 /* hardware has been reset, we need to reload some things */
4063 ixgbe_configure(adapter);
4065 ixgbe_up_complete(adapter);
4068 void ixgbe_reset(struct ixgbe_adapter *adapter)
4070 struct ixgbe_hw *hw = &adapter->hw;
4073 /* lock SFP init bit to prevent race conditions with the watchdog */
4074 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
4075 usleep_range(1000, 2000);
4077 /* clear all SFP and link config related flags while holding SFP_INIT */
4078 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
4079 IXGBE_FLAG2_SFP_NEEDS_RESET);
4080 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
4082 err = hw->mac.ops.init_hw(hw);
4085 case IXGBE_ERR_SFP_NOT_PRESENT:
4086 case IXGBE_ERR_SFP_NOT_SUPPORTED:
4088 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
4089 e_dev_err("master disable timed out\n");
4091 case IXGBE_ERR_EEPROM_VERSION:
4092 /* We are running on a pre-production device, log a warning */
4093 e_dev_warn("This device is a pre-production adapter/LOM. "
4094 "Please be aware there may be issues associated with "
4095 "your hardware. If you are experiencing problems "
4096 "please contact your Intel or hardware "
4097 "representative who provided you with this "
4101 e_dev_err("Hardware Error: %d\n", err);
4104 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
4106 /* reprogram the RAR[0] in case user changed it. */
4107 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
4112 * ixgbe_init_rx_page_offset - initialize page offset values for Rx buffers
4113 * @rx_ring: ring to setup
4115 * On many IA platforms the L1 cache has a critical stride of 4K, this
4116 * results in each receive buffer starting in the same cache set. To help
4117 * reduce the pressure on this cache set we can interleave the offsets so
4118 * that only every other buffer will be in the same cache set.
4120 static void ixgbe_init_rx_page_offset(struct ixgbe_ring *rx_ring)
4122 struct ixgbe_rx_buffer *rx_buffer = rx_ring->rx_buffer_info;
4125 for (i = 0; i < rx_ring->count; i += 2) {
4126 rx_buffer[0].page_offset = 0;
4127 rx_buffer[1].page_offset = ixgbe_rx_bufsz(rx_ring);
4128 rx_buffer = &rx_buffer[2];
4133 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4134 * @rx_ring: ring to free buffers from
4136 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4138 struct device *dev = rx_ring->dev;
4142 /* ring already cleared, nothing to do */
4143 if (!rx_ring->rx_buffer_info)
4146 /* Free all the Rx ring sk_buffs */
4147 for (i = 0; i < rx_ring->count; i++) {
4148 struct ixgbe_rx_buffer *rx_buffer;
4150 rx_buffer = &rx_ring->rx_buffer_info[i];
4151 if (rx_buffer->skb) {
4152 struct sk_buff *skb = rx_buffer->skb;
4153 if (IXGBE_CB(skb)->page_released) {
4156 ixgbe_rx_bufsz(rx_ring),
4158 IXGBE_CB(skb)->page_released = false;
4162 rx_buffer->skb = NULL;
4164 dma_unmap_page(dev, rx_buffer->dma,
4165 ixgbe_rx_pg_size(rx_ring),
4168 if (rx_buffer->page)
4169 __free_pages(rx_buffer->page,
4170 ixgbe_rx_pg_order(rx_ring));
4171 rx_buffer->page = NULL;
4174 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4175 memset(rx_ring->rx_buffer_info, 0, size);
4177 ixgbe_init_rx_page_offset(rx_ring);
4179 /* Zero out the descriptor ring */
4180 memset(rx_ring->desc, 0, rx_ring->size);
4182 rx_ring->next_to_alloc = 0;
4183 rx_ring->next_to_clean = 0;
4184 rx_ring->next_to_use = 0;
4188 * ixgbe_clean_tx_ring - Free Tx Buffers
4189 * @tx_ring: ring to be cleaned
4191 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
4193 struct ixgbe_tx_buffer *tx_buffer_info;
4197 /* ring already cleared, nothing to do */
4198 if (!tx_ring->tx_buffer_info)
4201 /* Free all the Tx ring sk_buffs */
4202 for (i = 0; i < tx_ring->count; i++) {
4203 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4204 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
4207 netdev_tx_reset_queue(txring_txq(tx_ring));
4209 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4210 memset(tx_ring->tx_buffer_info, 0, size);
4212 /* Zero out the descriptor ring */
4213 memset(tx_ring->desc, 0, tx_ring->size);
4215 tx_ring->next_to_use = 0;
4216 tx_ring->next_to_clean = 0;
4220 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4221 * @adapter: board private structure
4223 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4227 for (i = 0; i < adapter->num_rx_queues; i++)
4228 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4232 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4233 * @adapter: board private structure
4235 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4239 for (i = 0; i < adapter->num_tx_queues; i++)
4240 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4243 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
4245 struct hlist_node *node, *node2;
4246 struct ixgbe_fdir_filter *filter;
4248 spin_lock(&adapter->fdir_perfect_lock);
4250 hlist_for_each_entry_safe(filter, node, node2,
4251 &adapter->fdir_filter_list, fdir_node) {
4252 hlist_del(&filter->fdir_node);
4255 adapter->fdir_filter_count = 0;
4257 spin_unlock(&adapter->fdir_perfect_lock);
4260 void ixgbe_down(struct ixgbe_adapter *adapter)
4262 struct net_device *netdev = adapter->netdev;
4263 struct ixgbe_hw *hw = &adapter->hw;
4267 /* signal that we are down to the interrupt handler */
4268 set_bit(__IXGBE_DOWN, &adapter->state);
4270 /* disable receives */
4271 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4272 IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4274 /* disable all enabled rx queues */
4275 for (i = 0; i < adapter->num_rx_queues; i++)
4276 /* this call also flushes the previous write */
4277 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
4279 usleep_range(10000, 20000);
4281 netif_tx_stop_all_queues(netdev);
4283 /* call carrier off first to avoid false dev_watchdog timeouts */
4284 netif_carrier_off(netdev);
4285 netif_tx_disable(netdev);
4287 ixgbe_irq_disable(adapter);
4289 ixgbe_napi_disable_all(adapter);
4291 adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
4292 IXGBE_FLAG2_RESET_REQUESTED);
4293 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4295 del_timer_sync(&adapter->service_timer);
4297 if (adapter->num_vfs) {
4298 /* Clear EITR Select mapping */
4299 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
4301 /* Mark all the VFs as inactive */
4302 for (i = 0 ; i < adapter->num_vfs; i++)
4303 adapter->vfinfo[i].clear_to_send = false;
4305 /* ping all the active vfs to let them know we are going down */
4306 ixgbe_ping_all_vfs(adapter);
4308 /* Disable all VFTE/VFRE TX/RX */
4309 ixgbe_disable_tx_rx(adapter);
4312 /* disable transmits in the hardware now that interrupts are off */
4313 for (i = 0; i < adapter->num_tx_queues; i++) {
4314 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4315 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
4318 /* Disable the Tx DMA engine on 82599 and X540 */
4319 switch (hw->mac.type) {
4320 case ixgbe_mac_82599EB:
4321 case ixgbe_mac_X540:
4322 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4323 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4324 ~IXGBE_DMATXCTL_TE));
4330 if (!pci_channel_offline(adapter->pdev))
4331 ixgbe_reset(adapter);
4333 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
4334 if (hw->mac.ops.disable_tx_laser &&
4335 ((hw->phy.multispeed_fiber) ||
4336 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
4337 (hw->mac.type == ixgbe_mac_82599EB))))
4338 hw->mac.ops.disable_tx_laser(hw);
4340 ixgbe_clean_all_tx_rings(adapter);
4341 ixgbe_clean_all_rx_rings(adapter);
4343 #ifdef CONFIG_IXGBE_DCA
4344 /* since we reset the hardware DCA settings were cleared */
4345 ixgbe_setup_dca(adapter);
4350 * ixgbe_tx_timeout - Respond to a Tx Hang
4351 * @netdev: network interface device structure
4353 static void ixgbe_tx_timeout(struct net_device *netdev)
4355 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4357 /* Do the reset outside of interrupt context */
4358 ixgbe_tx_timeout_reset(adapter);
4362 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4363 * @adapter: board private structure to initialize
4365 * ixgbe_sw_init initializes the Adapter private data structure.
4366 * Fields are initialized based on PCI device information and
4367 * OS network device settings (MTU size).
4369 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4371 struct ixgbe_hw *hw = &adapter->hw;
4372 struct pci_dev *pdev = adapter->pdev;
4374 #ifdef CONFIG_IXGBE_DCB
4376 struct tc_configuration *tc;
4379 /* PCI config space info */
4381 hw->vendor_id = pdev->vendor;
4382 hw->device_id = pdev->device;
4383 hw->revision_id = pdev->revision;
4384 hw->subsystem_vendor_id = pdev->subsystem_vendor;
4385 hw->subsystem_device_id = pdev->subsystem_device;
4387 /* Set capability flags */
4388 rss = min_t(int, IXGBE_MAX_RSS_INDICES, num_online_cpus());
4389 adapter->ring_feature[RING_F_RSS].indices = rss;
4390 adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4391 switch (hw->mac.type) {
4392 case ixgbe_mac_82598EB:
4393 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4394 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4395 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
4397 case ixgbe_mac_X540:
4398 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4399 case ixgbe_mac_82599EB:
4400 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
4401 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4402 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4403 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4404 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4405 /* Flow Director hash filters enabled */
4406 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4407 adapter->atr_sample_rate = 20;
4408 adapter->ring_feature[RING_F_FDIR].indices =
4409 IXGBE_MAX_FDIR_INDICES;
4410 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
4412 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4413 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4414 adapter->ring_feature[RING_F_FCOE].indices = 0;
4415 #ifdef CONFIG_IXGBE_DCB
4416 /* Default traffic class to use for FCoE */
4417 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4419 #endif /* IXGBE_FCOE */
4425 /* n-tuple support exists, always init our spinlock */
4426 spin_lock_init(&adapter->fdir_perfect_lock);
4428 #ifdef CONFIG_IXGBE_DCB
4429 switch (hw->mac.type) {
4430 case ixgbe_mac_X540:
4431 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
4432 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
4435 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
4436 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
4440 /* Configure DCB traffic classes */
4441 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4442 tc = &adapter->dcb_cfg.tc_config[j];
4443 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4444 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4445 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4446 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4447 tc->dcb_pfc = pfc_disabled;
4450 /* Initialize default user to priority mapping, UPx->TC0 */
4451 tc = &adapter->dcb_cfg.tc_config[0];
4452 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
4453 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
4455 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4456 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4457 adapter->dcb_cfg.pfc_mode_enable = false;
4458 adapter->dcb_set_bitmap = 0x00;
4459 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
4460 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
4461 sizeof(adapter->temp_dcb_cfg));
4465 /* default flow control settings */
4466 hw->fc.requested_mode = ixgbe_fc_full;
4467 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
4468 ixgbe_pbthresh_setup(adapter);
4469 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4470 hw->fc.send_xon = true;
4471 hw->fc.disable_fc_autoneg = false;
4473 /* enable itr by default in dynamic mode */
4474 adapter->rx_itr_setting = 1;
4475 adapter->tx_itr_setting = 1;
4477 /* set default ring sizes */
4478 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4479 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4481 /* set default work limits */
4482 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
4484 /* initialize eeprom parameters */
4485 if (ixgbe_init_eeprom_params_generic(hw)) {
4486 e_dev_err("EEPROM initialization failed\n");
4490 set_bit(__IXGBE_DOWN, &adapter->state);
4496 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4497 * @tx_ring: tx descriptor ring (for a specific queue) to setup
4499 * Return 0 on success, negative on failure
4501 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4503 struct device *dev = tx_ring->dev;
4504 int orig_node = dev_to_node(dev);
4508 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4510 if (tx_ring->q_vector)
4511 numa_node = tx_ring->q_vector->numa_node;
4513 tx_ring->tx_buffer_info = vzalloc_node(size, numa_node);
4514 if (!tx_ring->tx_buffer_info)
4515 tx_ring->tx_buffer_info = vzalloc(size);
4516 if (!tx_ring->tx_buffer_info)
4519 /* round up to nearest 4K */
4520 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4521 tx_ring->size = ALIGN(tx_ring->size, 4096);
4523 set_dev_node(dev, numa_node);
4524 tx_ring->desc = dma_alloc_coherent(dev,
4528 set_dev_node(dev, orig_node);
4530 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4531 &tx_ring->dma, GFP_KERNEL);
4535 tx_ring->next_to_use = 0;
4536 tx_ring->next_to_clean = 0;
4540 vfree(tx_ring->tx_buffer_info);
4541 tx_ring->tx_buffer_info = NULL;
4542 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4547 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4548 * @adapter: board private structure
4550 * If this function returns with an error, then it's possible one or
4551 * more of the rings is populated (while the rest are not). It is the
4552 * callers duty to clean those orphaned rings.
4554 * Return 0 on success, negative on failure
4556 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4560 for (i = 0; i < adapter->num_tx_queues; i++) {
4561 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4564 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4572 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4573 * @rx_ring: rx descriptor ring (for a specific queue) to setup
4575 * Returns 0 on success, negative on failure
4577 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
4579 struct device *dev = rx_ring->dev;
4580 int orig_node = dev_to_node(dev);
4584 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4586 if (rx_ring->q_vector)
4587 numa_node = rx_ring->q_vector->numa_node;
4589 rx_ring->rx_buffer_info = vzalloc_node(size, numa_node);
4590 if (!rx_ring->rx_buffer_info)
4591 rx_ring->rx_buffer_info = vzalloc(size);
4592 if (!rx_ring->rx_buffer_info)
4595 /* Round up to nearest 4K */
4596 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4597 rx_ring->size = ALIGN(rx_ring->size, 4096);
4599 set_dev_node(dev, numa_node);
4600 rx_ring->desc = dma_alloc_coherent(dev,
4604 set_dev_node(dev, orig_node);
4606 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4607 &rx_ring->dma, GFP_KERNEL);
4611 rx_ring->next_to_clean = 0;
4612 rx_ring->next_to_use = 0;
4614 ixgbe_init_rx_page_offset(rx_ring);
4618 vfree(rx_ring->rx_buffer_info);
4619 rx_ring->rx_buffer_info = NULL;
4620 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4625 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4626 * @adapter: board private structure
4628 * If this function returns with an error, then it's possible one or
4629 * more of the rings is populated (while the rest are not). It is the
4630 * callers duty to clean those orphaned rings.
4632 * Return 0 on success, negative on failure
4634 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4638 for (i = 0; i < adapter->num_rx_queues; i++) {
4639 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
4642 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
4650 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4651 * @tx_ring: Tx descriptor ring for a specific queue
4653 * Free all transmit software resources
4655 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
4657 ixgbe_clean_tx_ring(tx_ring);
4659 vfree(tx_ring->tx_buffer_info);
4660 tx_ring->tx_buffer_info = NULL;
4662 /* if not set, then don't free */
4666 dma_free_coherent(tx_ring->dev, tx_ring->size,
4667 tx_ring->desc, tx_ring->dma);
4669 tx_ring->desc = NULL;
4673 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4674 * @adapter: board private structure
4676 * Free all transmit software resources
4678 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4682 for (i = 0; i < adapter->num_tx_queues; i++)
4683 if (adapter->tx_ring[i]->desc)
4684 ixgbe_free_tx_resources(adapter->tx_ring[i]);
4688 * ixgbe_free_rx_resources - Free Rx Resources
4689 * @rx_ring: ring to clean the resources from
4691 * Free all receive software resources
4693 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
4695 ixgbe_clean_rx_ring(rx_ring);
4697 vfree(rx_ring->rx_buffer_info);
4698 rx_ring->rx_buffer_info = NULL;
4700 /* if not set, then don't free */
4704 dma_free_coherent(rx_ring->dev, rx_ring->size,
4705 rx_ring->desc, rx_ring->dma);
4707 rx_ring->desc = NULL;
4711 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4712 * @adapter: board private structure
4714 * Free all receive software resources
4716 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4720 for (i = 0; i < adapter->num_rx_queues; i++)
4721 if (adapter->rx_ring[i]->desc)
4722 ixgbe_free_rx_resources(adapter->rx_ring[i]);
4726 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4727 * @netdev: network interface device structure
4728 * @new_mtu: new value for maximum frame size
4730 * Returns 0 on success, negative on failure
4732 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4734 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4735 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4737 /* MTU < 68 is an error and causes problems on some kernels */
4738 if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4742 * For 82599EB we cannot allow PF to change MTU greater than 1500
4743 * in SR-IOV mode as it may cause buffer overruns in guest VFs that
4744 * don't allocate and chain buffers correctly.
4746 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
4747 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
4748 (max_frame > MAXIMUM_ETHERNET_VLAN_SIZE))
4751 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4753 /* must set new MTU before calling down or up */
4754 netdev->mtu = new_mtu;
4756 if (netif_running(netdev))
4757 ixgbe_reinit_locked(adapter);
4763 * ixgbe_open - Called when a network interface is made active
4764 * @netdev: network interface device structure
4766 * Returns 0 on success, negative value on failure
4768 * The open entry point is called when a network interface is made
4769 * active by the system (IFF_UP). At this point all resources needed
4770 * for transmit and receive operations are allocated, the interrupt
4771 * handler is registered with the OS, the watchdog timer is started,
4772 * and the stack is notified that the interface is ready.
4774 static int ixgbe_open(struct net_device *netdev)
4776 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4779 /* disallow open during test */
4780 if (test_bit(__IXGBE_TESTING, &adapter->state))
4783 netif_carrier_off(netdev);
4785 /* allocate transmit descriptors */
4786 err = ixgbe_setup_all_tx_resources(adapter);
4790 /* allocate receive descriptors */
4791 err = ixgbe_setup_all_rx_resources(adapter);
4795 ixgbe_configure(adapter);
4797 err = ixgbe_request_irq(adapter);
4801 ixgbe_up_complete(adapter);
4807 ixgbe_free_all_rx_resources(adapter);
4809 ixgbe_free_all_tx_resources(adapter);
4810 ixgbe_reset(adapter);
4816 * ixgbe_close - Disables a network interface
4817 * @netdev: network interface device structure
4819 * Returns 0, this is not allowed to fail
4821 * The close entry point is called when an interface is de-activated
4822 * by the OS. The hardware is still under the drivers control, but
4823 * needs to be disabled. A global MAC reset is issued to stop the
4824 * hardware, and all transmit and receive resources are freed.
4826 static int ixgbe_close(struct net_device *netdev)
4828 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4830 ixgbe_down(adapter);
4831 ixgbe_free_irq(adapter);
4833 ixgbe_fdir_filter_exit(adapter);
4835 ixgbe_free_all_tx_resources(adapter);
4836 ixgbe_free_all_rx_resources(adapter);
4838 ixgbe_release_hw_control(adapter);
4844 static int ixgbe_resume(struct pci_dev *pdev)
4846 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4847 struct net_device *netdev = adapter->netdev;
4850 pci_set_power_state(pdev, PCI_D0);
4851 pci_restore_state(pdev);
4853 * pci_restore_state clears dev->state_saved so call
4854 * pci_save_state to restore it.
4856 pci_save_state(pdev);
4858 err = pci_enable_device_mem(pdev);
4860 e_dev_err("Cannot enable PCI device from suspend\n");
4863 pci_set_master(pdev);
4865 pci_wake_from_d3(pdev, false);
4868 err = ixgbe_init_interrupt_scheme(adapter);
4871 e_dev_err("Cannot initialize interrupts for device\n");
4875 ixgbe_reset(adapter);
4877 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4879 if (netif_running(netdev)) {
4880 err = ixgbe_open(netdev);
4885 netif_device_attach(netdev);
4889 #endif /* CONFIG_PM */
4891 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4893 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
4894 struct net_device *netdev = adapter->netdev;
4895 struct ixgbe_hw *hw = &adapter->hw;
4897 u32 wufc = adapter->wol;
4902 netif_device_detach(netdev);
4904 if (netif_running(netdev)) {
4906 ixgbe_down(adapter);
4907 ixgbe_free_irq(adapter);
4908 ixgbe_free_all_tx_resources(adapter);
4909 ixgbe_free_all_rx_resources(adapter);
4913 ixgbe_clear_interrupt_scheme(adapter);
4916 retval = pci_save_state(pdev);
4922 ixgbe_set_rx_mode(netdev);
4925 * enable the optics for both mult-speed fiber and
4926 * 82599 SFP+ fiber as we can WoL.
4928 if (hw->mac.ops.enable_tx_laser &&
4929 (hw->phy.multispeed_fiber ||
4930 (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber &&
4931 hw->mac.type == ixgbe_mac_82599EB)))
4932 hw->mac.ops.enable_tx_laser(hw);
4934 /* turn on all-multi mode if wake on multicast is enabled */
4935 if (wufc & IXGBE_WUFC_MC) {
4936 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4937 fctrl |= IXGBE_FCTRL_MPE;
4938 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4941 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4942 ctrl |= IXGBE_CTRL_GIO_DIS;
4943 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4945 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4947 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4948 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4951 switch (hw->mac.type) {
4952 case ixgbe_mac_82598EB:
4953 pci_wake_from_d3(pdev, false);
4955 case ixgbe_mac_82599EB:
4956 case ixgbe_mac_X540:
4957 pci_wake_from_d3(pdev, !!wufc);
4963 *enable_wake = !!wufc;
4965 ixgbe_release_hw_control(adapter);
4967 pci_disable_device(pdev);
4973 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4978 retval = __ixgbe_shutdown(pdev, &wake);
4983 pci_prepare_to_sleep(pdev);
4985 pci_wake_from_d3(pdev, false);
4986 pci_set_power_state(pdev, PCI_D3hot);
4991 #endif /* CONFIG_PM */
4993 static void ixgbe_shutdown(struct pci_dev *pdev)
4997 __ixgbe_shutdown(pdev, &wake);
4999 if (system_state == SYSTEM_POWER_OFF) {
5000 pci_wake_from_d3(pdev, wake);
5001 pci_set_power_state(pdev, PCI_D3hot);
5006 * ixgbe_update_stats - Update the board statistics counters.
5007 * @adapter: board private structure
5009 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5011 struct net_device *netdev = adapter->netdev;
5012 struct ixgbe_hw *hw = &adapter->hw;
5013 struct ixgbe_hw_stats *hwstats = &adapter->stats;
5015 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5016 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5017 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5018 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
5020 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
5022 u64 fcoe_noddp_counts_sum = 0, fcoe_noddp_ext_buff_counts_sum = 0;
5023 #endif /* IXGBE_FCOE */
5025 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5026 test_bit(__IXGBE_RESETTING, &adapter->state))
5029 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5032 for (i = 0; i < adapter->num_rx_queues; i++) {
5033 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5034 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5036 adapter->rsc_total_count = rsc_count;
5037 adapter->rsc_total_flush = rsc_flush;
5040 for (i = 0; i < adapter->num_rx_queues; i++) {
5041 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5042 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5043 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5044 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5045 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
5046 bytes += rx_ring->stats.bytes;
5047 packets += rx_ring->stats.packets;
5049 adapter->non_eop_descs = non_eop_descs;
5050 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5051 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5052 adapter->hw_csum_rx_error = hw_csum_rx_error;
5053 netdev->stats.rx_bytes = bytes;
5054 netdev->stats.rx_packets = packets;
5058 /* gather some stats to the adapter struct that are per queue */
5059 for (i = 0; i < adapter->num_tx_queues; i++) {
5060 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5061 restart_queue += tx_ring->tx_stats.restart_queue;
5062 tx_busy += tx_ring->tx_stats.tx_busy;
5063 bytes += tx_ring->stats.bytes;
5064 packets += tx_ring->stats.packets;
5066 adapter->restart_queue = restart_queue;
5067 adapter->tx_busy = tx_busy;
5068 netdev->stats.tx_bytes = bytes;
5069 netdev->stats.tx_packets = packets;
5071 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5073 /* 8 register reads */
5074 for (i = 0; i < 8; i++) {
5075 /* for packet buffers not used, the register should read 0 */
5076 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5078 hwstats->mpc[i] += mpc;
5079 total_mpc += hwstats->mpc[i];
5080 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5081 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5082 switch (hw->mac.type) {
5083 case ixgbe_mac_82598EB:
5084 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5085 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5086 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5087 hwstats->pxonrxc[i] +=
5088 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5090 case ixgbe_mac_82599EB:
5091 case ixgbe_mac_X540:
5092 hwstats->pxonrxc[i] +=
5093 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5100 /*16 register reads */
5101 for (i = 0; i < 16; i++) {
5102 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5103 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5104 if ((hw->mac.type == ixgbe_mac_82599EB) ||
5105 (hw->mac.type == ixgbe_mac_X540)) {
5106 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
5107 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
5108 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
5109 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
5113 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5114 /* work around hardware counting issue */
5115 hwstats->gprc -= missed_rx;
5117 ixgbe_update_xoff_received(adapter);
5119 /* 82598 hardware only has a 32 bit counter in the high register */
5120 switch (hw->mac.type) {
5121 case ixgbe_mac_82598EB:
5122 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5123 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5124 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5125 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5127 case ixgbe_mac_X540:
5128 /* OS2BMC stats are X540 only*/
5129 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
5130 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
5131 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
5132 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
5133 case ixgbe_mac_82599EB:
5134 for (i = 0; i < 16; i++)
5135 adapter->hw_rx_no_dma_resources +=
5136 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5137 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5138 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5139 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5140 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5141 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5142 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5143 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5144 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5145 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5147 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5148 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5149 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5150 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5151 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5152 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5153 /* Add up per cpu counters for total ddp aloc fail */
5154 if (fcoe->pcpu_noddp && fcoe->pcpu_noddp_ext_buff) {
5155 for_each_possible_cpu(cpu) {
5156 fcoe_noddp_counts_sum +=
5157 *per_cpu_ptr(fcoe->pcpu_noddp, cpu);
5158 fcoe_noddp_ext_buff_counts_sum +=
5160 pcpu_noddp_ext_buff, cpu);
5163 hwstats->fcoe_noddp = fcoe_noddp_counts_sum;
5164 hwstats->fcoe_noddp_ext_buff = fcoe_noddp_ext_buff_counts_sum;
5165 #endif /* IXGBE_FCOE */
5170 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5171 hwstats->bprc += bprc;
5172 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5173 if (hw->mac.type == ixgbe_mac_82598EB)
5174 hwstats->mprc -= bprc;
5175 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5176 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5177 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5178 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5179 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5180 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5181 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5182 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5183 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5184 hwstats->lxontxc += lxon;
5185 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5186 hwstats->lxofftxc += lxoff;
5187 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5188 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5190 * 82598 errata - tx of flow control packets is included in tx counters
5192 xon_off_tot = lxon + lxoff;
5193 hwstats->gptc -= xon_off_tot;
5194 hwstats->mptc -= xon_off_tot;
5195 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5196 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5197 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5198 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5199 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5200 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5201 hwstats->ptc64 -= xon_off_tot;
5202 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5203 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5204 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5205 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5206 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5207 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5209 /* Fill out the OS statistics structure */
5210 netdev->stats.multicast = hwstats->mprc;
5213 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5214 netdev->stats.rx_dropped = 0;
5215 netdev->stats.rx_length_errors = hwstats->rlec;
5216 netdev->stats.rx_crc_errors = hwstats->crcerrs;
5217 netdev->stats.rx_missed_errors = total_mpc;
5221 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
5222 * @adapter: pointer to the device adapter structure
5224 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
5226 struct ixgbe_hw *hw = &adapter->hw;
5229 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
5232 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5234 /* if interface is down do nothing */
5235 if (test_bit(__IXGBE_DOWN, &adapter->state))
5238 /* do nothing if we are not using signature filters */
5239 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
5242 adapter->fdir_overflow++;
5244 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5245 for (i = 0; i < adapter->num_tx_queues; i++)
5246 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5247 &(adapter->tx_ring[i]->state));
5248 /* re-enable flow director interrupts */
5249 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
5251 e_err(probe, "failed to finish FDIR re-initialization, "
5252 "ignored adding FDIR ATR filters\n");
5257 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
5258 * @adapter: pointer to the device adapter structure
5260 * This function serves two purposes. First it strobes the interrupt lines
5261 * in order to make certain interrupts are occurring. Secondly it sets the
5262 * bits needed to check for TX hangs. As a result we should immediately
5263 * determine if a hang has occurred.
5265 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
5267 struct ixgbe_hw *hw = &adapter->hw;
5271 /* If we're down or resetting, just bail */
5272 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5273 test_bit(__IXGBE_RESETTING, &adapter->state))
5276 /* Force detection of hung controller */
5277 if (netif_carrier_ok(adapter->netdev)) {
5278 for (i = 0; i < adapter->num_tx_queues; i++)
5279 set_check_for_tx_hang(adapter->tx_ring[i]);
5282 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5284 * for legacy and MSI interrupts don't set any bits
5285 * that are enabled for EIAM, because this operation
5286 * would set *both* EIMS and EICS for any bit in EIAM
5288 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5289 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5291 /* get one bit for every active tx/rx interrupt vector */
5292 for (i = 0; i < adapter->num_q_vectors; i++) {
5293 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5294 if (qv->rx.ring || qv->tx.ring)
5295 eics |= ((u64)1 << i);
5299 /* Cause software interrupt to ensure rings are cleaned */
5300 ixgbe_irq_rearm_queues(adapter, eics);
5305 * ixgbe_watchdog_update_link - update the link status
5306 * @adapter: pointer to the device adapter structure
5307 * @link_speed: pointer to a u32 to store the link_speed
5309 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
5311 struct ixgbe_hw *hw = &adapter->hw;
5312 u32 link_speed = adapter->link_speed;
5313 bool link_up = adapter->link_up;
5314 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
5316 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5319 if (hw->mac.ops.check_link) {
5320 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5322 /* always assume link is up, if no check link function */
5323 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
5327 if (adapter->ixgbe_ieee_pfc)
5328 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
5330 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
5331 hw->mac.ops.fc_enable(hw);
5332 ixgbe_set_rx_drop_en(adapter);
5336 time_after(jiffies, (adapter->link_check_timeout +
5337 IXGBE_TRY_LINK_TIMEOUT))) {
5338 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5339 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5340 IXGBE_WRITE_FLUSH(hw);
5343 adapter->link_up = link_up;
5344 adapter->link_speed = link_speed;
5348 * ixgbe_watchdog_link_is_up - update netif_carrier status and
5349 * print link up message
5350 * @adapter: pointer to the device adapter structure
5352 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
5354 struct net_device *netdev = adapter->netdev;
5355 struct ixgbe_hw *hw = &adapter->hw;
5356 u32 link_speed = adapter->link_speed;
5357 bool flow_rx, flow_tx;
5359 /* only continue if link was previously down */
5360 if (netif_carrier_ok(netdev))
5363 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
5365 switch (hw->mac.type) {
5366 case ixgbe_mac_82598EB: {
5367 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5368 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5369 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5370 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5373 case ixgbe_mac_X540:
5374 case ixgbe_mac_82599EB: {
5375 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5376 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5377 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5378 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5387 #ifdef CONFIG_IXGBE_PTP
5388 ixgbe_ptp_start_cyclecounter(adapter);
5391 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5392 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5394 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5396 (link_speed == IXGBE_LINK_SPEED_100_FULL ?
5399 ((flow_rx && flow_tx) ? "RX/TX" :
5401 (flow_tx ? "TX" : "None"))));
5403 netif_carrier_on(netdev);
5404 ixgbe_check_vf_rate_limit(adapter);
5408 * ixgbe_watchdog_link_is_down - update netif_carrier status and
5409 * print link down message
5410 * @adapter: pointer to the adapter structure
5412 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
5414 struct net_device *netdev = adapter->netdev;
5415 struct ixgbe_hw *hw = &adapter->hw;
5417 adapter->link_up = false;
5418 adapter->link_speed = 0;
5420 /* only continue if link was up previously */
5421 if (!netif_carrier_ok(netdev))
5424 /* poll for SFP+ cable when link is down */
5425 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
5426 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5428 #ifdef CONFIG_IXGBE_PTP
5429 ixgbe_ptp_start_cyclecounter(adapter);
5432 e_info(drv, "NIC Link is Down\n");
5433 netif_carrier_off(netdev);
5437 * ixgbe_watchdog_flush_tx - flush queues on link down
5438 * @adapter: pointer to the device adapter structure
5440 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
5443 int some_tx_pending = 0;
5445 if (!netif_carrier_ok(adapter->netdev)) {
5446 for (i = 0; i < adapter->num_tx_queues; i++) {
5447 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5448 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5449 some_tx_pending = 1;
5454 if (some_tx_pending) {
5455 /* We've lost link, so the controller stops DMA,
5456 * but we've got queued Tx work that's never going
5457 * to get done, so reset controller to flush Tx.
5458 * (Do the reset outside of interrupt context).
5460 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
5465 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
5469 /* Do not perform spoof check for 82598 */
5470 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5473 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
5476 * ssvpc register is cleared on read, if zero then no
5477 * spoofed packets in the last interval.
5482 e_warn(drv, "%d Spoofed packets detected\n", ssvpc);
5486 * ixgbe_watchdog_subtask - check and bring link up
5487 * @adapter: pointer to the device adapter structure
5489 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
5491 /* if interface is down do nothing */
5492 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5493 test_bit(__IXGBE_RESETTING, &adapter->state))
5496 ixgbe_watchdog_update_link(adapter);
5498 if (adapter->link_up)
5499 ixgbe_watchdog_link_is_up(adapter);
5501 ixgbe_watchdog_link_is_down(adapter);
5503 ixgbe_spoof_check(adapter);
5504 ixgbe_update_stats(adapter);
5506 ixgbe_watchdog_flush_tx(adapter);
5510 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
5511 * @adapter: the ixgbe adapter structure
5513 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
5515 struct ixgbe_hw *hw = &adapter->hw;
5518 /* not searching for SFP so there is nothing to do here */
5519 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
5520 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5523 /* someone else is in init, wait until next service event */
5524 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5527 err = hw->phy.ops.identify_sfp(hw);
5528 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5531 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
5532 /* If no cable is present, then we need to reset
5533 * the next time we find a good cable. */
5534 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5541 /* exit if reset not needed */
5542 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
5545 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
5548 * A module may be identified correctly, but the EEPROM may not have
5549 * support for that module. setup_sfp() will fail in that case, so
5550 * we should not allow that module to load.
5552 if (hw->mac.type == ixgbe_mac_82598EB)
5553 err = hw->phy.ops.reset(hw);
5555 err = hw->mac.ops.setup_sfp(hw);
5557 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
5560 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
5561 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5564 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5566 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
5567 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
5568 e_dev_err("failed to initialize because an unsupported "
5569 "SFP+ module type was detected.\n");
5570 e_dev_err("Reload the driver after installing a "
5571 "supported module.\n");
5572 unregister_netdev(adapter->netdev);
5577 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
5578 * @adapter: the ixgbe adapter structure
5580 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
5582 struct ixgbe_hw *hw = &adapter->hw;
5586 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
5589 /* someone else is in init, wait until next service event */
5590 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5593 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5595 autoneg = hw->phy.autoneg_advertised;
5596 if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5597 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5598 if (hw->mac.ops.setup_link)
5599 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5601 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5602 adapter->link_check_timeout = jiffies;
5603 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5606 #ifdef CONFIG_PCI_IOV
5607 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
5610 struct ixgbe_hw *hw = &adapter->hw;
5611 struct net_device *netdev = adapter->netdev;
5615 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
5616 if (gpc) /* If incrementing then no need for the check below */
5619 * Check to see if a bad DMA write target from an errant or
5620 * malicious VF has caused a PCIe error. If so then we can
5621 * issue a VFLR to the offending VF(s) and then resume without
5622 * requesting a full slot reset.
5625 for (vf = 0; vf < adapter->num_vfs; vf++) {
5626 ciaa = (vf << 16) | 0x80000000;
5627 /* 32 bit read so align, we really want status at offset 6 */
5628 ciaa |= PCI_COMMAND;
5629 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5630 ciad = IXGBE_READ_REG(hw, IXGBE_CIAD_82599);
5632 /* disable debug mode asap after reading data */
5633 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5634 /* Get the upper 16 bits which will be the PCI status reg */
5636 if (ciad & PCI_STATUS_REC_MASTER_ABORT) {
5637 netdev_err(netdev, "VF %d Hung DMA\n", vf);
5639 ciaa = (vf << 16) | 0x80000000;
5641 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5642 ciad = 0x00008000; /* VFLR */
5643 IXGBE_WRITE_REG(hw, IXGBE_CIAD_82599, ciad);
5645 IXGBE_WRITE_REG(hw, IXGBE_CIAA_82599, ciaa);
5652 * ixgbe_service_timer - Timer Call-back
5653 * @data: pointer to adapter cast into an unsigned long
5655 static void ixgbe_service_timer(unsigned long data)
5657 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5658 unsigned long next_event_offset;
5661 /* poll faster when waiting for link */
5662 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
5663 next_event_offset = HZ / 10;
5665 next_event_offset = HZ * 2;
5667 #ifdef CONFIG_PCI_IOV
5669 * don't bother with SR-IOV VF DMA hang check if there are
5670 * no VFs or the link is down
5672 if (!adapter->num_vfs ||
5673 (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
5674 goto normal_timer_service;
5676 /* If we have VFs allocated then we must check for DMA hangs */
5677 ixgbe_check_for_bad_vf(adapter);
5678 next_event_offset = HZ / 50;
5679 adapter->timer_event_accumulator++;
5681 if (adapter->timer_event_accumulator >= 100)
5682 adapter->timer_event_accumulator = 0;
5686 normal_timer_service:
5688 /* Reset the timer */
5689 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
5692 ixgbe_service_event_schedule(adapter);
5695 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
5697 if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
5700 adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
5702 /* If we're already down or resetting, just bail */
5703 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5704 test_bit(__IXGBE_RESETTING, &adapter->state))
5707 ixgbe_dump(adapter);
5708 netdev_err(adapter->netdev, "Reset adapter\n");
5709 adapter->tx_timeout_count++;
5711 ixgbe_reinit_locked(adapter);
5715 * ixgbe_service_task - manages and runs subtasks
5716 * @work: pointer to work_struct containing our data
5718 static void ixgbe_service_task(struct work_struct *work)
5720 struct ixgbe_adapter *adapter = container_of(work,
5721 struct ixgbe_adapter,
5724 ixgbe_reset_subtask(adapter);
5725 ixgbe_sfp_detection_subtask(adapter);
5726 ixgbe_sfp_link_config_subtask(adapter);
5727 ixgbe_check_overtemp_subtask(adapter);
5728 ixgbe_watchdog_subtask(adapter);
5729 ixgbe_fdir_reinit_subtask(adapter);
5730 ixgbe_check_hang_subtask(adapter);
5731 #ifdef CONFIG_IXGBE_PTP
5732 ixgbe_ptp_overflow_check(adapter);
5735 ixgbe_service_event_complete(adapter);
5738 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
5739 struct ixgbe_tx_buffer *first,
5742 struct sk_buff *skb = first->skb;
5743 u32 vlan_macip_lens, type_tucmd;
5744 u32 mss_l4len_idx, l4len;
5746 if (!skb_is_gso(skb))
5749 if (skb_header_cloned(skb)) {
5750 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5755 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5756 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
5758 if (first->protocol == __constant_htons(ETH_P_IP)) {
5759 struct iphdr *iph = ip_hdr(skb);
5762 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5766 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5767 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5768 IXGBE_TX_FLAGS_CSUM |
5769 IXGBE_TX_FLAGS_IPV4;
5770 } else if (skb_is_gso_v6(skb)) {
5771 ipv6_hdr(skb)->payload_len = 0;
5772 tcp_hdr(skb)->check =
5773 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5774 &ipv6_hdr(skb)->daddr,
5776 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
5777 IXGBE_TX_FLAGS_CSUM;
5780 /* compute header lengths */
5781 l4len = tcp_hdrlen(skb);
5782 *hdr_len = skb_transport_offset(skb) + l4len;
5784 /* update gso size and bytecount with header size */
5785 first->gso_segs = skb_shinfo(skb)->gso_segs;
5786 first->bytecount += (first->gso_segs - 1) * *hdr_len;
5788 /* mss_l4len_id: use 1 as index for TSO */
5789 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
5790 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
5791 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
5793 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
5794 vlan_macip_lens = skb_network_header_len(skb);
5795 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5796 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5798 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
5804 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
5805 struct ixgbe_tx_buffer *first)
5807 struct sk_buff *skb = first->skb;
5808 u32 vlan_macip_lens = 0;
5809 u32 mss_l4len_idx = 0;
5812 if (skb->ip_summed != CHECKSUM_PARTIAL) {
5813 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
5814 !(first->tx_flags & IXGBE_TX_FLAGS_TXSW))
5818 switch (first->protocol) {
5819 case __constant_htons(ETH_P_IP):
5820 vlan_macip_lens |= skb_network_header_len(skb);
5821 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
5822 l4_hdr = ip_hdr(skb)->protocol;
5824 case __constant_htons(ETH_P_IPV6):
5825 vlan_macip_lens |= skb_network_header_len(skb);
5826 l4_hdr = ipv6_hdr(skb)->nexthdr;
5829 if (unlikely(net_ratelimit())) {
5830 dev_warn(tx_ring->dev,
5831 "partial checksum but proto=%x!\n",
5839 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5840 mss_l4len_idx = tcp_hdrlen(skb) <<
5841 IXGBE_ADVTXD_L4LEN_SHIFT;
5844 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5845 mss_l4len_idx = sizeof(struct sctphdr) <<
5846 IXGBE_ADVTXD_L4LEN_SHIFT;
5849 mss_l4len_idx = sizeof(struct udphdr) <<
5850 IXGBE_ADVTXD_L4LEN_SHIFT;
5853 if (unlikely(net_ratelimit())) {
5854 dev_warn(tx_ring->dev,
5855 "partial checksum but l4 proto=%x!\n",
5861 /* update TX checksum flag */
5862 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
5865 /* vlan_macip_lens: MACLEN, VLAN tag */
5866 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
5867 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
5869 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
5870 type_tucmd, mss_l4len_idx);
5873 static __le32 ixgbe_tx_cmd_type(u32 tx_flags)
5875 /* set type for advanced descriptor with frame checksum insertion */
5876 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
5877 IXGBE_ADVTXD_DCMD_IFCS |
5878 IXGBE_ADVTXD_DCMD_DEXT);
5880 /* set HW vlan bit if vlan is present */
5881 if (tx_flags & IXGBE_TX_FLAGS_HW_VLAN)
5882 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
5884 #ifdef CONFIG_IXGBE_PTP
5885 if (tx_flags & IXGBE_TX_FLAGS_TSTAMP)
5886 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_MAC_TSTAMP);
5889 /* set segmentation enable bits for TSO/FSO */
5891 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FSO))
5893 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5895 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
5900 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
5901 u32 tx_flags, unsigned int paylen)
5903 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
5905 /* enable L4 checksum for TSO and TX checksum offload */
5906 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5907 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
5909 /* enble IPv4 checksum for TSO */
5910 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5911 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
5913 /* use index 1 context for TSO/FSO/FCOE */
5915 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_FCOE))
5917 if (tx_flags & IXGBE_TX_FLAGS_TSO)
5919 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
5922 * Check Context must be set if Tx switch is enabled, which it
5923 * always is for case where virtual functions are running
5926 if (tx_flags & (IXGBE_TX_FLAGS_TXSW | IXGBE_TX_FLAGS_FCOE))
5928 if (tx_flags & IXGBE_TX_FLAGS_TXSW)
5930 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
5932 tx_desc->read.olinfo_status = olinfo_status;
5935 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
5938 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
5939 struct ixgbe_tx_buffer *first,
5943 struct sk_buff *skb = first->skb;
5944 struct ixgbe_tx_buffer *tx_buffer;
5945 union ixgbe_adv_tx_desc *tx_desc;
5946 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
5947 unsigned int data_len = skb->data_len;
5948 unsigned int size = skb_headlen(skb);
5949 unsigned int paylen = skb->len - hdr_len;
5950 u32 tx_flags = first->tx_flags;
5952 u16 i = tx_ring->next_to_use;
5954 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5956 ixgbe_tx_olinfo_status(tx_desc, tx_flags, paylen);
5957 cmd_type = ixgbe_tx_cmd_type(tx_flags);
5960 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5961 if (data_len < sizeof(struct fcoe_crc_eof)) {
5962 size -= sizeof(struct fcoe_crc_eof) - data_len;
5965 data_len -= sizeof(struct fcoe_crc_eof);
5970 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5971 if (dma_mapping_error(tx_ring->dev, dma))
5974 /* record length, and DMA address */
5975 dma_unmap_len_set(first, len, size);
5976 dma_unmap_addr_set(first, dma, dma);
5978 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5981 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
5982 tx_desc->read.cmd_type_len =
5983 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
5987 if (i == tx_ring->count) {
5988 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5992 dma += IXGBE_MAX_DATA_PER_TXD;
5993 size -= IXGBE_MAX_DATA_PER_TXD;
5995 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5996 tx_desc->read.olinfo_status = 0;
5999 if (likely(!data_len))
6002 if (unlikely(skb->no_fcs))
6003 cmd_type &= ~(cpu_to_le32(IXGBE_ADVTXD_DCMD_IFCS));
6004 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
6008 if (i == tx_ring->count) {
6009 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6014 size = min_t(unsigned int, data_len, skb_frag_size(frag));
6016 size = skb_frag_size(frag);
6020 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
6022 if (dma_mapping_error(tx_ring->dev, dma))
6025 tx_buffer = &tx_ring->tx_buffer_info[i];
6026 dma_unmap_len_set(tx_buffer, len, size);
6027 dma_unmap_addr_set(tx_buffer, dma, dma);
6029 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6030 tx_desc->read.olinfo_status = 0;
6035 /* write last descriptor with RS and EOP bits */
6036 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
6037 tx_desc->read.cmd_type_len = cmd_type;
6039 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6041 /* set the timestamp */
6042 first->time_stamp = jiffies;
6045 * Force memory writes to complete before letting h/w know there
6046 * are new descriptors to fetch. (Only applicable for weak-ordered
6047 * memory model archs, such as IA-64).
6049 * We also need this memory barrier to make certain all of the
6050 * status bits have been updated before next_to_watch is written.
6054 /* set next_to_watch value indicating a packet is present */
6055 first->next_to_watch = tx_desc;
6058 if (i == tx_ring->count)
6061 tx_ring->next_to_use = i;
6063 /* notify HW of packet */
6064 writel(i, tx_ring->tail);
6068 dev_err(tx_ring->dev, "TX DMA map failed\n");
6070 /* clear dma mappings for failed tx_buffer_info map */
6072 tx_buffer = &tx_ring->tx_buffer_info[i];
6073 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
6074 if (tx_buffer == first)
6081 tx_ring->next_to_use = i;
6084 static void ixgbe_atr(struct ixgbe_ring *ring,
6085 struct ixgbe_tx_buffer *first)
6087 struct ixgbe_q_vector *q_vector = ring->q_vector;
6088 union ixgbe_atr_hash_dword input = { .dword = 0 };
6089 union ixgbe_atr_hash_dword common = { .dword = 0 };
6091 unsigned char *network;
6093 struct ipv6hdr *ipv6;
6098 /* if ring doesn't have a interrupt vector, cannot perform ATR */
6102 /* do nothing if sampling is disabled */
6103 if (!ring->atr_sample_rate)
6108 /* snag network header to get L4 type and address */
6109 hdr.network = skb_network_header(first->skb);
6111 /* Currently only IPv4/IPv6 with TCP is supported */
6112 if ((first->protocol != __constant_htons(ETH_P_IPV6) ||
6113 hdr.ipv6->nexthdr != IPPROTO_TCP) &&
6114 (first->protocol != __constant_htons(ETH_P_IP) ||
6115 hdr.ipv4->protocol != IPPROTO_TCP))
6118 th = tcp_hdr(first->skb);
6120 /* skip this packet since it is invalid or the socket is closing */
6124 /* sample on all syn packets or once every atr sample count */
6125 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
6128 /* reset sample count */
6129 ring->atr_count = 0;
6131 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
6134 * src and dst are inverted, think how the receiver sees them
6136 * The input is broken into two sections, a non-compressed section
6137 * containing vm_pool, vlan_id, and flow_type. The rest of the data
6138 * is XORed together and stored in the compressed dword.
6140 input.formatted.vlan_id = vlan_id;
6143 * since src port and flex bytes occupy the same word XOR them together
6144 * and write the value to source port portion of compressed dword
6146 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
6147 common.port.src ^= th->dest ^ __constant_htons(ETH_P_8021Q);
6149 common.port.src ^= th->dest ^ first->protocol;
6150 common.port.dst ^= th->source;
6152 if (first->protocol == __constant_htons(ETH_P_IP)) {
6153 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
6154 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
6156 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
6157 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
6158 hdr.ipv6->saddr.s6_addr32[1] ^
6159 hdr.ipv6->saddr.s6_addr32[2] ^
6160 hdr.ipv6->saddr.s6_addr32[3] ^
6161 hdr.ipv6->daddr.s6_addr32[0] ^
6162 hdr.ipv6->daddr.s6_addr32[1] ^
6163 hdr.ipv6->daddr.s6_addr32[2] ^
6164 hdr.ipv6->daddr.s6_addr32[3];
6167 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6168 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
6169 input, common, ring->queue_index);
6172 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6174 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6175 /* Herbert's original patch had:
6176 * smp_mb__after_netif_stop_queue();
6177 * but since that doesn't exist yet, just open code it. */
6180 /* We need to check again in a case another CPU has just
6181 * made room available. */
6182 if (likely(ixgbe_desc_unused(tx_ring) < size))
6185 /* A reprieve! - use start_queue because it doesn't call schedule */
6186 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6187 ++tx_ring->tx_stats.restart_queue;
6191 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
6193 if (likely(ixgbe_desc_unused(tx_ring) >= size))
6195 return __ixgbe_maybe_stop_tx(tx_ring, size);
6198 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6200 struct ixgbe_adapter *adapter = netdev_priv(dev);
6201 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
6204 __be16 protocol = vlan_get_protocol(skb);
6206 if (((protocol == htons(ETH_P_FCOE)) ||
6207 (protocol == htons(ETH_P_FIP))) &&
6208 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6209 txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6210 txq += adapter->ring_feature[RING_F_FCOE].mask;
6215 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6216 while (unlikely(txq >= dev->real_num_tx_queues))
6217 txq -= dev->real_num_tx_queues;
6221 return skb_tx_hash(dev, skb);
6224 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6225 struct ixgbe_adapter *adapter,
6226 struct ixgbe_ring *tx_ring)
6228 struct ixgbe_tx_buffer *first;
6231 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6234 u16 count = TXD_USE_COUNT(skb_headlen(skb));
6235 __be16 protocol = skb->protocol;
6239 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
6240 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
6241 * + 2 desc gap to keep tail from touching head,
6242 * + 1 desc for context descriptor,
6243 * otherwise try next time
6245 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
6246 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6247 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6249 count += skb_shinfo(skb)->nr_frags;
6251 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
6252 tx_ring->tx_stats.tx_busy++;
6253 return NETDEV_TX_BUSY;
6256 /* record the location of the first descriptor for this packet */
6257 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6259 first->bytecount = skb->len;
6260 first->gso_segs = 1;
6262 /* if we have a HW VLAN tag being added default to the HW one */
6263 if (vlan_tx_tag_present(skb)) {
6264 tx_flags |= vlan_tx_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
6265 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6266 /* else if it is a SW VLAN check the next protocol and store the tag */
6267 } else if (protocol == __constant_htons(ETH_P_8021Q)) {
6268 struct vlan_hdr *vhdr, _vhdr;
6269 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
6273 protocol = vhdr->h_vlan_encapsulated_proto;
6274 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
6275 IXGBE_TX_FLAGS_VLAN_SHIFT;
6276 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
6279 skb_tx_timestamp(skb);
6281 #ifdef CONFIG_IXGBE_PTP
6282 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6283 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6284 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
6288 #ifdef CONFIG_PCI_IOV
6290 * Use the l2switch_enable flag - would be false if the DMA
6291 * Tx switch had been disabled.
6293 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6294 tx_flags |= IXGBE_TX_FLAGS_TXSW;
6297 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
6298 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
6299 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
6300 (skb->priority != TC_PRIO_CONTROL))) {
6301 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6302 tx_flags |= (skb->priority & 0x7) <<
6303 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
6304 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
6305 struct vlan_ethhdr *vhdr;
6306 if (skb_header_cloned(skb) &&
6307 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
6309 vhdr = (struct vlan_ethhdr *)skb->data;
6310 vhdr->h_vlan_TCI = htons(tx_flags >>
6311 IXGBE_TX_FLAGS_VLAN_SHIFT);
6313 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
6317 /* record initial flags and protocol */
6318 first->tx_flags = tx_flags;
6319 first->protocol = protocol;
6322 /* setup tx offload for FCoE */
6323 if ((protocol == __constant_htons(ETH_P_FCOE)) &&
6324 (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)) {
6325 tso = ixgbe_fso(tx_ring, first, &hdr_len);
6332 #endif /* IXGBE_FCOE */
6333 tso = ixgbe_tso(tx_ring, first, &hdr_len);
6337 ixgbe_tx_csum(tx_ring, first);
6339 /* add the ATR filter if ATR is on */
6340 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
6341 ixgbe_atr(tx_ring, first);
6345 #endif /* IXGBE_FCOE */
6346 ixgbe_tx_map(tx_ring, first, hdr_len);
6348 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6350 return NETDEV_TX_OK;
6353 dev_kfree_skb_any(first->skb);
6356 return NETDEV_TX_OK;
6359 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
6360 struct net_device *netdev)
6362 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6363 struct ixgbe_ring *tx_ring;
6366 * The minimum packet size for olinfo paylen is 17 so pad the skb
6367 * in order to meet this minimum size requirement.
6369 if (unlikely(skb->len < 17)) {
6370 if (skb_pad(skb, 17 - skb->len))
6371 return NETDEV_TX_OK;
6375 tx_ring = adapter->tx_ring[skb->queue_mapping];
6376 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6380 * ixgbe_set_mac - Change the Ethernet Address of the NIC
6381 * @netdev: network interface device structure
6382 * @p: pointer to an address structure
6384 * Returns 0 on success, negative on failure
6386 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6388 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6389 struct ixgbe_hw *hw = &adapter->hw;
6390 struct sockaddr *addr = p;
6392 if (!is_valid_ether_addr(addr->sa_data))
6393 return -EADDRNOTAVAIL;
6395 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6396 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6398 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6405 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6407 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6408 struct ixgbe_hw *hw = &adapter->hw;
6412 if (prtad != hw->phy.mdio.prtad)
6414 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6420 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6421 u16 addr, u16 value)
6423 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6424 struct ixgbe_hw *hw = &adapter->hw;
6426 if (prtad != hw->phy.mdio.prtad)
6428 return hw->phy.ops.write_reg(hw, addr, devad, value);
6431 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6433 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6436 #ifdef CONFIG_IXGBE_PTP
6438 return ixgbe_ptp_hwtstamp_ioctl(adapter, req, cmd);
6441 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6446 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6448 * @netdev: network interface device structure
6450 * Returns non-zero on failure
6452 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6455 struct ixgbe_adapter *adapter = netdev_priv(dev);
6456 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6458 if (is_valid_ether_addr(mac->san_addr)) {
6460 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6467 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6469 * @netdev: network interface device structure
6471 * Returns non-zero on failure
6473 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6476 struct ixgbe_adapter *adapter = netdev_priv(dev);
6477 struct ixgbe_mac_info *mac = &adapter->hw.mac;
6479 if (is_valid_ether_addr(mac->san_addr)) {
6481 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6487 #ifdef CONFIG_NET_POLL_CONTROLLER
6489 * Polling 'interrupt' - used by things like netconsole to send skbs
6490 * without having to re-enable interrupts. It's not called while
6491 * the interrupt routine is executing.
6493 static void ixgbe_netpoll(struct net_device *netdev)
6495 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6498 /* if interface is down do nothing */
6499 if (test_bit(__IXGBE_DOWN, &adapter->state))
6502 adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6503 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6504 for (i = 0; i < adapter->num_q_vectors; i++)
6505 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
6507 ixgbe_intr(adapter->pdev->irq, netdev);
6509 adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6513 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6514 struct rtnl_link_stats64 *stats)
6516 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6520 for (i = 0; i < adapter->num_rx_queues; i++) {
6521 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6527 start = u64_stats_fetch_begin_bh(&ring->syncp);
6528 packets = ring->stats.packets;
6529 bytes = ring->stats.bytes;
6530 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6531 stats->rx_packets += packets;
6532 stats->rx_bytes += bytes;
6536 for (i = 0; i < adapter->num_tx_queues; i++) {
6537 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
6543 start = u64_stats_fetch_begin_bh(&ring->syncp);
6544 packets = ring->stats.packets;
6545 bytes = ring->stats.bytes;
6546 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6547 stats->tx_packets += packets;
6548 stats->tx_bytes += bytes;
6552 /* following stats updated by ixgbe_watchdog_task() */
6553 stats->multicast = netdev->stats.multicast;
6554 stats->rx_errors = netdev->stats.rx_errors;
6555 stats->rx_length_errors = netdev->stats.rx_length_errors;
6556 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
6557 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6561 #ifdef CONFIG_IXGBE_DCB
6563 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
6564 * @adapter: pointer to ixgbe_adapter
6565 * @tc: number of traffic classes currently enabled
6567 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
6568 * 802.1Q priority maps to a packet buffer that exists.
6570 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
6572 struct ixgbe_hw *hw = &adapter->hw;
6576 /* 82598 have a static priority to TC mapping that can not
6577 * be changed so no validation is needed.
6579 if (hw->mac.type == ixgbe_mac_82598EB)
6582 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
6585 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
6586 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
6588 /* If up2tc is out of bounds default to zero */
6590 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
6594 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
6600 * ixgbe_setup_tc - configure net_device for multiple traffic classes
6602 * @netdev: net device to configure
6603 * @tc: number of traffic classes to enable
6605 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
6607 struct ixgbe_adapter *adapter = netdev_priv(dev);
6608 struct ixgbe_hw *hw = &adapter->hw;
6610 /* Multiple traffic classes requires multiple queues */
6611 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6612 e_err(drv, "Enable failed, needs MSI-X\n");
6616 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6617 e_err(drv, "Enable failed, SR-IOV enabled\n");
6621 /* Hardware supports up to 8 traffic classes */
6622 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs ||
6623 (hw->mac.type == ixgbe_mac_82598EB &&
6624 tc < MAX_TRAFFIC_CLASS))
6627 /* Hardware has to reinitialize queues and interrupts to
6628 * match packet buffer alignment. Unfortunately, the
6629 * hardware is not flexible enough to do this dynamically.
6631 if (netif_running(dev))
6633 ixgbe_clear_interrupt_scheme(adapter);
6636 netdev_set_num_tc(dev, tc);
6637 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
6638 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6640 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
6641 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
6642 adapter->hw.fc.requested_mode = ixgbe_fc_none;
6645 netdev_reset_tc(dev);
6646 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
6647 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
6649 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6650 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6652 adapter->temp_dcb_cfg.pfc_mode_enable = false;
6653 adapter->dcb_cfg.pfc_mode_enable = false;
6656 ixgbe_init_interrupt_scheme(adapter);
6657 ixgbe_validate_rtr(adapter, tc);
6658 if (netif_running(dev))
6664 #endif /* CONFIG_IXGBE_DCB */
6665 void ixgbe_do_reset(struct net_device *netdev)
6667 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6669 if (netif_running(netdev))
6670 ixgbe_reinit_locked(adapter);
6672 ixgbe_reset(adapter);
6675 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
6676 netdev_features_t features)
6678 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6680 /* return error if RXHASH is being enabled when RSS is not supported */
6681 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
6682 features &= ~NETIF_F_RXHASH;
6684 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
6685 if (!(features & NETIF_F_RXCSUM))
6686 features &= ~NETIF_F_LRO;
6688 /* Turn off LRO if not RSC capable */
6689 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
6690 features &= ~NETIF_F_LRO;
6695 static int ixgbe_set_features(struct net_device *netdev,
6696 netdev_features_t features)
6698 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6699 netdev_features_t changed = netdev->features ^ features;
6700 bool need_reset = false;
6702 /* Make sure RSC matches LRO, reset if change */
6703 if (!(features & NETIF_F_LRO)) {
6704 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6706 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
6707 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
6708 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
6709 if (adapter->rx_itr_setting == 1 ||
6710 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
6711 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
6713 } else if ((changed ^ features) & NETIF_F_LRO) {
6714 e_info(probe, "rx-usecs set too low, "
6720 * Check if Flow Director n-tuple support was enabled or disabled. If
6721 * the state changed, we need to reset.
6723 if (!(features & NETIF_F_NTUPLE)) {
6724 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
6725 /* turn off Flow Director, set ATR and reset */
6726 if ((adapter->flags & IXGBE_FLAG_RSS_ENABLED) &&
6727 !(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
6728 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
6731 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6732 } else if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
6733 /* turn off ATR, enable perfect filters and reset */
6734 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
6735 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
6739 if (features & NETIF_F_HW_VLAN_RX)
6740 ixgbe_vlan_strip_enable(adapter);
6742 ixgbe_vlan_strip_disable(adapter);
6744 if (changed & NETIF_F_RXALL)
6747 netdev->features = features;
6749 ixgbe_do_reset(netdev);
6754 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm,
6755 struct net_device *dev,
6756 unsigned char *addr,
6759 struct ixgbe_adapter *adapter = netdev_priv(dev);
6760 int err = -EOPNOTSUPP;
6762 if (ndm->ndm_state & NUD_PERMANENT) {
6763 pr_info("%s: FDB only supports static addresses\n",
6768 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6769 if (is_unicast_ether_addr(addr))
6770 err = dev_uc_add_excl(dev, addr);
6771 else if (is_multicast_ether_addr(addr))
6772 err = dev_mc_add_excl(dev, addr);
6777 /* Only return duplicate errors if NLM_F_EXCL is set */
6778 if (err == -EEXIST && !(flags & NLM_F_EXCL))
6784 static int ixgbe_ndo_fdb_del(struct ndmsg *ndm,
6785 struct net_device *dev,
6786 unsigned char *addr)
6788 struct ixgbe_adapter *adapter = netdev_priv(dev);
6789 int err = -EOPNOTSUPP;
6791 if (ndm->ndm_state & NUD_PERMANENT) {
6792 pr_info("%s: FDB only supports static addresses\n",
6797 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6798 if (is_unicast_ether_addr(addr))
6799 err = dev_uc_del(dev, addr);
6800 else if (is_multicast_ether_addr(addr))
6801 err = dev_mc_del(dev, addr);
6809 static int ixgbe_ndo_fdb_dump(struct sk_buff *skb,
6810 struct netlink_callback *cb,
6811 struct net_device *dev,
6814 struct ixgbe_adapter *adapter = netdev_priv(dev);
6816 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6817 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
6822 static const struct net_device_ops ixgbe_netdev_ops = {
6823 .ndo_open = ixgbe_open,
6824 .ndo_stop = ixgbe_close,
6825 .ndo_start_xmit = ixgbe_xmit_frame,
6826 .ndo_select_queue = ixgbe_select_queue,
6827 .ndo_set_rx_mode = ixgbe_set_rx_mode,
6828 .ndo_validate_addr = eth_validate_addr,
6829 .ndo_set_mac_address = ixgbe_set_mac,
6830 .ndo_change_mtu = ixgbe_change_mtu,
6831 .ndo_tx_timeout = ixgbe_tx_timeout,
6832 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
6833 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
6834 .ndo_do_ioctl = ixgbe_ioctl,
6835 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
6836 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
6837 .ndo_set_vf_tx_rate = ixgbe_ndo_set_vf_bw,
6838 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
6839 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
6840 .ndo_get_stats64 = ixgbe_get_stats64,
6841 #ifdef CONFIG_IXGBE_DCB
6842 .ndo_setup_tc = ixgbe_setup_tc,
6844 #ifdef CONFIG_NET_POLL_CONTROLLER
6845 .ndo_poll_controller = ixgbe_netpoll,
6848 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6849 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
6850 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6851 .ndo_fcoe_enable = ixgbe_fcoe_enable,
6852 .ndo_fcoe_disable = ixgbe_fcoe_disable,
6853 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6854 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
6855 #endif /* IXGBE_FCOE */
6856 .ndo_set_features = ixgbe_set_features,
6857 .ndo_fix_features = ixgbe_fix_features,
6858 .ndo_fdb_add = ixgbe_ndo_fdb_add,
6859 .ndo_fdb_del = ixgbe_ndo_fdb_del,
6860 .ndo_fdb_dump = ixgbe_ndo_fdb_dump,
6863 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6864 const struct ixgbe_info *ii)
6866 #ifdef CONFIG_PCI_IOV
6867 struct ixgbe_hw *hw = &adapter->hw;
6869 if (hw->mac.type == ixgbe_mac_82598EB)
6872 /* The 82599 supports up to 64 VFs per physical function
6873 * but this implementation limits allocation to 63 so that
6874 * basic networking resources are still available to the
6875 * physical function. If the user requests greater thn
6876 * 63 VFs then it is an error - reset to default of zero.
6878 adapter->num_vfs = (max_vfs > 63) ? 0 : max_vfs;
6879 ixgbe_enable_sriov(adapter, ii);
6880 #endif /* CONFIG_PCI_IOV */
6884 * ixgbe_wol_supported - Check whether device supports WoL
6885 * @hw: hw specific details
6886 * @device_id: the device ID
6887 * @subdev_id: the subsystem device ID
6889 * This function is used by probe and ethtool to determine
6890 * which devices have WoL support
6893 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
6896 struct ixgbe_hw *hw = &adapter->hw;
6897 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
6898 int is_wol_supported = 0;
6900 switch (device_id) {
6901 case IXGBE_DEV_ID_82599_SFP:
6902 /* Only these subdevices could supports WOL */
6903 switch (subdevice_id) {
6904 case IXGBE_SUBDEV_ID_82599_560FLR:
6905 /* only support first port */
6906 if (hw->bus.func != 0)
6908 case IXGBE_SUBDEV_ID_82599_SFP:
6909 is_wol_supported = 1;
6913 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
6914 /* All except this subdevice support WOL */
6915 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
6916 is_wol_supported = 1;
6918 case IXGBE_DEV_ID_82599_KX4:
6919 is_wol_supported = 1;
6921 case IXGBE_DEV_ID_X540T:
6922 /* check eeprom to see if enabled wol */
6923 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
6924 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
6925 (hw->bus.func == 0))) {
6926 is_wol_supported = 1;
6931 return is_wol_supported;
6935 * ixgbe_probe - Device Initialization Routine
6936 * @pdev: PCI device information struct
6937 * @ent: entry in ixgbe_pci_tbl
6939 * Returns 0 on success, negative on failure
6941 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6942 * The OS initialization, configuring of the adapter private structure,
6943 * and a hardware reset occur.
6945 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6946 const struct pci_device_id *ent)
6948 struct net_device *netdev;
6949 struct ixgbe_adapter *adapter = NULL;
6950 struct ixgbe_hw *hw;
6951 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6952 static int cards_found;
6953 int i, err, pci_using_dac;
6954 u8 part_str[IXGBE_PBANUM_LENGTH];
6955 unsigned int indices = num_possible_cpus();
6961 /* Catch broken hardware that put the wrong VF device ID in
6962 * the PCIe SR-IOV capability.
6964 if (pdev->is_virtfn) {
6965 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6966 pci_name(pdev), pdev->vendor, pdev->device);
6970 err = pci_enable_device_mem(pdev);
6974 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6975 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6978 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6980 err = dma_set_coherent_mask(&pdev->dev,
6984 "No usable DMA configuration, aborting\n");
6991 err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6992 IORESOURCE_MEM), ixgbe_driver_name);
6995 "pci_request_selected_regions failed 0x%x\n", err);
6999 pci_enable_pcie_error_reporting(pdev);
7001 pci_set_master(pdev);
7002 pci_save_state(pdev);
7004 #ifdef CONFIG_IXGBE_DCB
7005 indices *= MAX_TRAFFIC_CLASS;
7008 if (ii->mac == ixgbe_mac_82598EB)
7009 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7011 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7014 indices += min_t(unsigned int, num_possible_cpus(),
7015 IXGBE_MAX_FCOE_INDICES);
7017 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7020 goto err_alloc_etherdev;
7023 SET_NETDEV_DEV(netdev, &pdev->dev);
7025 adapter = netdev_priv(netdev);
7026 pci_set_drvdata(pdev, adapter);
7028 adapter->netdev = netdev;
7029 adapter->pdev = pdev;
7032 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
7034 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7035 pci_resource_len(pdev, 0));
7041 for (i = 1; i <= 5; i++) {
7042 if (pci_resource_len(pdev, i) == 0)
7046 netdev->netdev_ops = &ixgbe_netdev_ops;
7047 ixgbe_set_ethtool_ops(netdev);
7048 netdev->watchdog_timeo = 5 * HZ;
7049 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
7051 adapter->bd_number = cards_found;
7054 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7055 hw->mac.type = ii->mac;
7058 memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7059 eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7060 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7061 if (!(eec & (1 << 8)))
7062 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7065 memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7066 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7067 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7068 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7069 hw->phy.mdio.mmds = 0;
7070 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7071 hw->phy.mdio.dev = netdev;
7072 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7073 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7075 ii->get_invariants(hw);
7077 /* setup the private structure */
7078 err = ixgbe_sw_init(adapter);
7082 /* Make it possible the adapter to be woken up via WOL */
7083 switch (adapter->hw.mac.type) {
7084 case ixgbe_mac_82599EB:
7085 case ixgbe_mac_X540:
7086 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7093 * If there is a fan on this device and it has failed log the
7096 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7097 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7098 if (esdp & IXGBE_ESDP_SDP1)
7099 e_crit(probe, "Fan has stopped, replace the adapter\n");
7102 if (allow_unsupported_sfp)
7103 hw->allow_unsupported_sfp = allow_unsupported_sfp;
7105 /* reset_hw fills in the perm_addr as well */
7106 hw->phy.reset_if_overtemp = true;
7107 err = hw->mac.ops.reset_hw(hw);
7108 hw->phy.reset_if_overtemp = false;
7109 if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7110 hw->mac.type == ixgbe_mac_82598EB) {
7112 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7113 e_dev_err("failed to load because an unsupported SFP+ "
7114 "module type was detected.\n");
7115 e_dev_err("Reload the driver after installing a supported "
7119 e_dev_err("HW Init failed: %d\n", err);
7123 ixgbe_probe_vf(adapter, ii);
7125 netdev->features = NETIF_F_SG |
7128 NETIF_F_HW_VLAN_TX |
7129 NETIF_F_HW_VLAN_RX |
7130 NETIF_F_HW_VLAN_FILTER |
7136 netdev->hw_features = netdev->features;
7138 switch (adapter->hw.mac.type) {
7139 case ixgbe_mac_82599EB:
7140 case ixgbe_mac_X540:
7141 netdev->features |= NETIF_F_SCTP_CSUM;
7142 netdev->hw_features |= NETIF_F_SCTP_CSUM |
7149 netdev->hw_features |= NETIF_F_RXALL;
7151 netdev->vlan_features |= NETIF_F_TSO;
7152 netdev->vlan_features |= NETIF_F_TSO6;
7153 netdev->vlan_features |= NETIF_F_IP_CSUM;
7154 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7155 netdev->vlan_features |= NETIF_F_SG;
7157 netdev->priv_flags |= IFF_UNICAST_FLT;
7158 netdev->priv_flags |= IFF_SUPP_NOFCS;
7160 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7161 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7162 IXGBE_FLAG_DCB_ENABLED);
7164 #ifdef CONFIG_IXGBE_DCB
7165 netdev->dcbnl_ops = &dcbnl_ops;
7169 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7170 if (hw->mac.ops.get_device_caps) {
7171 hw->mac.ops.get_device_caps(hw, &device_caps);
7172 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7173 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7176 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7177 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7178 netdev->vlan_features |= NETIF_F_FSO;
7179 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7181 #endif /* IXGBE_FCOE */
7182 if (pci_using_dac) {
7183 netdev->features |= NETIF_F_HIGHDMA;
7184 netdev->vlan_features |= NETIF_F_HIGHDMA;
7187 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
7188 netdev->hw_features |= NETIF_F_LRO;
7189 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7190 netdev->features |= NETIF_F_LRO;
7192 /* make sure the EEPROM is good */
7193 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7194 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7199 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7200 memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7202 if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7203 e_dev_err("invalid MAC address\n");
7208 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
7209 (unsigned long) adapter);
7211 INIT_WORK(&adapter->service_task, ixgbe_service_task);
7212 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
7214 err = ixgbe_init_interrupt_scheme(adapter);
7218 if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED)) {
7219 netdev->hw_features &= ~NETIF_F_RXHASH;
7220 netdev->features &= ~NETIF_F_RXHASH;
7223 /* WOL not supported for all devices */
7225 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
7226 if (ixgbe_wol_supported(adapter, pdev->device, pdev->subsystem_device))
7227 adapter->wol = IXGBE_WUFC_MAG;
7229 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7231 #ifdef CONFIG_IXGBE_PTP
7232 ixgbe_ptp_init(adapter);
7233 #endif /* CONFIG_IXGBE_PTP*/
7235 /* save off EEPROM version number */
7236 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
7237 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
7239 /* pick up the PCI bus settings for reporting later */
7240 hw->mac.ops.get_bus_info(hw);
7242 /* print bus type/speed/width info */
7243 e_dev_info("(PCI Express:%s:%s) %pM\n",
7244 (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0GT/s" :
7245 hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5GT/s" :
7247 (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7248 hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7249 hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7253 err = ixgbe_read_pba_string_generic(hw, part_str, IXGBE_PBANUM_LENGTH);
7255 strncpy(part_str, "Unknown", IXGBE_PBANUM_LENGTH);
7256 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7257 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
7258 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7261 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
7262 hw->mac.type, hw->phy.type, part_str);
7264 if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7265 e_dev_warn("PCI-Express bandwidth available for this card is "
7266 "not sufficient for optimal performance.\n");
7267 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7271 /* reset the hardware with the new settings */
7272 err = hw->mac.ops.start_hw(hw);
7273 if (err == IXGBE_ERR_EEPROM_VERSION) {
7274 /* We are running on a pre-production device, log a warning */
7275 e_dev_warn("This device is a pre-production adapter/LOM. "
7276 "Please be aware there may be issues associated "
7277 "with your hardware. If you are experiencing "
7278 "problems please contact your Intel or hardware "
7279 "representative who provided you with this "
7282 strcpy(netdev->name, "eth%d");
7283 err = register_netdev(netdev);
7287 /* power down the optics for multispeed fiber and 82599 SFP+ fiber */
7288 if (hw->mac.ops.disable_tx_laser &&
7289 ((hw->phy.multispeed_fiber) ||
7290 ((hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) &&
7291 (hw->mac.type == ixgbe_mac_82599EB))))
7292 hw->mac.ops.disable_tx_laser(hw);
7294 /* carrier off reporting is important to ethtool even BEFORE open */
7295 netif_carrier_off(netdev);
7297 #ifdef CONFIG_IXGBE_DCA
7298 if (dca_add_requester(&pdev->dev) == 0) {
7299 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7300 ixgbe_setup_dca(adapter);
7303 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7304 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7305 for (i = 0; i < adapter->num_vfs; i++)
7306 ixgbe_vf_configuration(pdev, (i | 0x10000000));
7309 /* firmware requires driver version to be 0xFFFFFFFF
7310 * since os does not support feature
7312 if (hw->mac.ops.set_fw_drv_ver)
7313 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
7316 /* add san mac addr to netdev */
7317 ixgbe_add_sanmac_netdev(netdev);
7319 e_dev_info("%s\n", ixgbe_default_device_descr);
7322 #ifdef CONFIG_IXGBE_HWMON
7323 if (ixgbe_sysfs_init(adapter))
7324 e_err(probe, "failed to allocate sysfs resources\n");
7325 #endif /* CONFIG_IXGBE_HWMON */
7330 ixgbe_release_hw_control(adapter);
7331 ixgbe_clear_interrupt_scheme(adapter);
7333 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7334 ixgbe_disable_sriov(adapter);
7335 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7336 iounmap(hw->hw_addr);
7338 free_netdev(netdev);
7340 pci_release_selected_regions(pdev,
7341 pci_select_bars(pdev, IORESOURCE_MEM));
7344 pci_disable_device(pdev);
7349 * ixgbe_remove - Device Removal Routine
7350 * @pdev: PCI device information struct
7352 * ixgbe_remove is called by the PCI subsystem to alert the driver
7353 * that it should release a PCI device. The could be caused by a
7354 * Hot-Plug event, or because the driver is going to be removed from
7357 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7359 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7360 struct net_device *netdev = adapter->netdev;
7362 set_bit(__IXGBE_DOWN, &adapter->state);
7363 cancel_work_sync(&adapter->service_task);
7365 #ifdef CONFIG_IXGBE_PTP
7366 ixgbe_ptp_stop(adapter);
7369 #ifdef CONFIG_IXGBE_DCA
7370 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7371 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7372 dca_remove_requester(&pdev->dev);
7373 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7377 #ifdef CONFIG_IXGBE_HWMON
7378 ixgbe_sysfs_exit(adapter);
7379 #endif /* CONFIG_IXGBE_HWMON */
7382 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7383 ixgbe_cleanup_fcoe(adapter);
7385 #endif /* IXGBE_FCOE */
7387 /* remove the added san mac */
7388 ixgbe_del_sanmac_netdev(netdev);
7390 if (netdev->reg_state == NETREG_REGISTERED)
7391 unregister_netdev(netdev);
7393 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7394 if (!(ixgbe_check_vf_assignment(adapter)))
7395 ixgbe_disable_sriov(adapter);
7397 e_dev_warn("Unloading driver while VFs are assigned "
7398 "- VFs will not be deallocated\n");
7401 ixgbe_clear_interrupt_scheme(adapter);
7403 ixgbe_release_hw_control(adapter);
7406 kfree(adapter->ixgbe_ieee_pfc);
7407 kfree(adapter->ixgbe_ieee_ets);
7410 iounmap(adapter->hw.hw_addr);
7411 pci_release_selected_regions(pdev, pci_select_bars(pdev,
7414 e_dev_info("complete\n");
7416 free_netdev(netdev);
7418 pci_disable_pcie_error_reporting(pdev);
7420 pci_disable_device(pdev);
7424 * ixgbe_io_error_detected - called when PCI error is detected
7425 * @pdev: Pointer to PCI device
7426 * @state: The current pci connection state
7428 * This function is called after a PCI bus error affecting
7429 * this device has been detected.
7431 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7432 pci_channel_state_t state)
7434 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7435 struct net_device *netdev = adapter->netdev;
7437 #ifdef CONFIG_PCI_IOV
7438 struct pci_dev *bdev, *vfdev;
7439 u32 dw0, dw1, dw2, dw3;
7441 u16 req_id, pf_func;
7443 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7444 adapter->num_vfs == 0)
7445 goto skip_bad_vf_detection;
7447 bdev = pdev->bus->self;
7448 while (bdev && (bdev->pcie_type != PCI_EXP_TYPE_ROOT_PORT))
7449 bdev = bdev->bus->self;
7452 goto skip_bad_vf_detection;
7454 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
7456 goto skip_bad_vf_detection;
7458 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG, &dw0);
7459 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 4, &dw1);
7460 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 8, &dw2);
7461 pci_read_config_dword(bdev, pos + PCI_ERR_HEADER_LOG + 12, &dw3);
7464 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
7465 if (!(req_id & 0x0080))
7466 goto skip_bad_vf_detection;
7468 pf_func = req_id & 0x01;
7469 if ((pf_func & 1) == (pdev->devfn & 1)) {
7470 unsigned int device_id;
7472 vf = (req_id & 0x7F) >> 1;
7473 e_dev_err("VF %d has caused a PCIe error\n", vf);
7474 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
7475 "%8.8x\tdw3: %8.8x\n",
7476 dw0, dw1, dw2, dw3);
7477 switch (adapter->hw.mac.type) {
7478 case ixgbe_mac_82599EB:
7479 device_id = IXGBE_82599_VF_DEVICE_ID;
7481 case ixgbe_mac_X540:
7482 device_id = IXGBE_X540_VF_DEVICE_ID;
7489 /* Find the pci device of the offending VF */
7490 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID, device_id, NULL);
7492 if (vfdev->devfn == (req_id & 0xFF))
7494 vfdev = pci_get_device(IXGBE_INTEL_VENDOR_ID,
7498 * There's a slim chance the VF could have been hot plugged,
7499 * so if it is no longer present we don't need to issue the
7500 * VFLR. Just clean up the AER in that case.
7503 e_dev_err("Issuing VFLR to VF %d\n", vf);
7504 pci_write_config_dword(vfdev, 0xA8, 0x00008000);
7507 pci_cleanup_aer_uncorrect_error_status(pdev);
7511 * Even though the error may have occurred on the other port
7512 * we still need to increment the vf error reference count for
7513 * both ports because the I/O resume function will be called
7516 adapter->vferr_refcount++;
7518 return PCI_ERS_RESULT_RECOVERED;
7520 skip_bad_vf_detection:
7521 #endif /* CONFIG_PCI_IOV */
7522 netif_device_detach(netdev);
7524 if (state == pci_channel_io_perm_failure)
7525 return PCI_ERS_RESULT_DISCONNECT;
7527 if (netif_running(netdev))
7528 ixgbe_down(adapter);
7529 pci_disable_device(pdev);
7531 /* Request a slot reset. */
7532 return PCI_ERS_RESULT_NEED_RESET;
7536 * ixgbe_io_slot_reset - called after the pci bus has been reset.
7537 * @pdev: Pointer to PCI device
7539 * Restart the card from scratch, as if from a cold-boot.
7541 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7543 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7544 pci_ers_result_t result;
7547 if (pci_enable_device_mem(pdev)) {
7548 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7549 result = PCI_ERS_RESULT_DISCONNECT;
7551 pci_set_master(pdev);
7552 pci_restore_state(pdev);
7553 pci_save_state(pdev);
7555 pci_wake_from_d3(pdev, false);
7557 ixgbe_reset(adapter);
7558 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7559 result = PCI_ERS_RESULT_RECOVERED;
7562 err = pci_cleanup_aer_uncorrect_error_status(pdev);
7564 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7565 "failed 0x%0x\n", err);
7566 /* non-fatal, continue */
7573 * ixgbe_io_resume - called when traffic can start flowing again.
7574 * @pdev: Pointer to PCI device
7576 * This callback is called when the error recovery driver tells us that
7577 * its OK to resume normal operation.
7579 static void ixgbe_io_resume(struct pci_dev *pdev)
7581 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7582 struct net_device *netdev = adapter->netdev;
7584 #ifdef CONFIG_PCI_IOV
7585 if (adapter->vferr_refcount) {
7586 e_info(drv, "Resuming after VF err\n");
7587 adapter->vferr_refcount--;
7592 if (netif_running(netdev))
7595 netif_device_attach(netdev);
7598 static struct pci_error_handlers ixgbe_err_handler = {
7599 .error_detected = ixgbe_io_error_detected,
7600 .slot_reset = ixgbe_io_slot_reset,
7601 .resume = ixgbe_io_resume,
7604 static struct pci_driver ixgbe_driver = {
7605 .name = ixgbe_driver_name,
7606 .id_table = ixgbe_pci_tbl,
7607 .probe = ixgbe_probe,
7608 .remove = __devexit_p(ixgbe_remove),
7610 .suspend = ixgbe_suspend,
7611 .resume = ixgbe_resume,
7613 .shutdown = ixgbe_shutdown,
7614 .err_handler = &ixgbe_err_handler
7618 * ixgbe_init_module - Driver Registration Routine
7620 * ixgbe_init_module is the first routine called when the driver is
7621 * loaded. All it does is register with the PCI subsystem.
7623 static int __init ixgbe_init_module(void)
7626 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7627 pr_info("%s\n", ixgbe_copyright);
7629 #ifdef CONFIG_IXGBE_DCA
7630 dca_register_notify(&dca_notifier);
7633 ret = pci_register_driver(&ixgbe_driver);
7637 module_init(ixgbe_init_module);
7640 * ixgbe_exit_module - Driver Exit Cleanup Routine
7642 * ixgbe_exit_module is called just before the driver is removed
7645 static void __exit ixgbe_exit_module(void)
7647 #ifdef CONFIG_IXGBE_DCA
7648 dca_unregister_notify(&dca_notifier);
7650 pci_unregister_driver(&ixgbe_driver);
7651 rcu_barrier(); /* Wait for completion of call_rcu()'s */
7654 #ifdef CONFIG_IXGBE_DCA
7655 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7660 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7661 __ixgbe_notify_dca);
7663 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7666 #endif /* CONFIG_IXGBE_DCA */
7668 module_exit(ixgbe_exit_module);